Deep Learning Wikipedia

Deep Learning more at http://ml.memect.com

Contents
1

Artiﬁcial neural network

1

1.1

Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

1

1.2

History . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

2

1.2.1

Improvements since 2006 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

2

Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

3

1.3.1

Network function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

3

1.3.2

Learning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

4

1.3.3

Learning paradigms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

4

1.3.4

Learning algorithms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

5

1.4

Employing artiﬁcial neural networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

5

1.5

Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

6

1.5.1

Real-life applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

6

1.5.2

Neural networks and neuroscience . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

6

1.6

Neural network software . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

6

1.7

Types of artiﬁcial neural networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

7

1.8

Theoretical properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

7

1.8.1

Computational power . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

7

1.8.2

Capacity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

7

1.8.3

Convergence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

7

1.8.4

Generalization and statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

7

Controversies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

8

1.9.1

Training issues . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

8

1.9.2

Hardware issues . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

8

1.9.3

Practical counterexamples to criticisms . . . . . . . . . . . . . . . . . . . . . . . . . . . .

8

1.9.4

Hybrid approaches . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

8

1.10 Gallery . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

9

1.11 See also . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

9

1.12 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

9

1.13 Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

11

1.14 External links . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

12

Deep learning

13

2.1

13

1.3

1.9

2

Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . i ii

CONTENTS
2.1.1

Deﬁnitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

13

2.1.2

Fundamental concepts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

14

2.2

History . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

14

2.3

Deep learning in artiﬁcial neural networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

15

2.4

Deep learning architectures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

16

2.4.1

Deep neural networks

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

16

2.4.2

Issues with deep neural networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

17

2.4.3

Deep belief networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

17

2.4.4

Convolutional neural networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

18

2.4.5

Convolutional Deep Belief Networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

18

2.4.6

Deep Boltzmann Machines . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

18

2.4.7

Stacked (Denoising) Auto-Encoders . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

19

2.4.8

Deep Stacking Networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

19

2.4.9

Tensor Deep Stacking Networks (T-DSN) . . . . . . . . . . . . . . . . . . . . . . . . . .

20

2.4.10 Spike-and-Slab RBMs (ssRBMs) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

20

2.4.11 Compound Hierarchical-Deep Models . . . . . . . . . . . . . . . . . . . . . . . . . . . .

20

2.4.12 Deep Coding Networks

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

21

2.4.13 Deep Kernel Machines . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

21

2.4.14 Deep Q-Networks

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

21

Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

21

2.5.1

Automatic speech recognition

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

22

2.5.2

Image recognition

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

22

2.5.3

Natural language processing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

23

2.5.4

Drug discovery and toxicology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

23

2.5.5

Customer relationship management

. . . . . . . . . . . . . . . . . . . . . . . . . . . . .

23

2.6

Deep learning in the human brain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

23

2.7

Commercial activity

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

24

2.8

Criticism and comment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

24

2.9

Deep learning software libraries

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

25

2.10 See also . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

25

2.11 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

25

2.12 External links . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

30

Feature learning

32

3.1

Supervised feature learning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

32

3.1.1

Supervised dictionary learning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

32

3.1.2

Neural networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

32

2.5

3

3.2

Unsupervised feature learning

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

33

3.2.1

K-means clustering . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

33

3.2.2

Principal component analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

33

3.2.3

Local linear embedding

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

33

3.2.4

Independent component analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

34

CONTENTS
3.2.5
3.3

iii
Unsupervised dictionary learning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

34

3.3.1

Restricted Boltzmann machine . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

34

3.3.2

Autoencoder . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

34

3.4

See also . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

34

3.5
4

Multilayer/Deep architectures

34

References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

35

Unsupervised learning

36

4.1

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

36

4.2

See also . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

36

4.3

Notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

37

4.4
5

Method of moments

Further reading . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

37
38

5.1

See also . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

38

5.2

References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

38

5.3
6

Generative model

Sources . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

38
39

6.1

Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

39

6.2

Encoding and decoding . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

39

6.3

Coding schemes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

39

6.3.1

Rate coding . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

40

6.3.2

Temporal coding . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

41

6.3.3

Population coding . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

43

6.3.4

Sparse coding . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

44

6.4

See also . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

45

6.5

References

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

45

6.6
7

Neural coding

Further reading . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

47
48

7.1

See also . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

48

7.2
8

Word embedding
References

48

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

49

8.1

Training algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

49

8.2

See also . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

49

8.3

References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

49

8.4
9

Deep belief network

External links . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

50

Convolutional neural network

51

9.1

Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

51

9.2

History

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

51

9.3

Details

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

52

iv

CONTENTS
9.3.1

Backpropagation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

52

9.3.2

Diﬀerent types of layers

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

52

Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

53

9.4.1

Image recognition

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

53

9.4.2

Video analysis

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

53

9.4.3

Natural Language Processing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

53

9.4.4

Playing Go . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

53

9.5

Fine-tuning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

53

9.6

Common libraries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

54

9.7

See also . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

54

9.8

References

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

54

9.9

External links . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

55

9.4

10 Restricted Boltzmann machine

56

10.1 Structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

56

10.1.1 Relation to other models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

57

10.2 Training algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

57

10.3 See also . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

58

10.4 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

58

10.5 External links . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

58

11 Recurrent neural network

59

11.1 Architectures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

59

11.1.1 Fully recurrent network . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

59

11.1.2 Hopﬁeld network . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

59

11.1.3 Elman networks and Jordan networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

59

11.1.4 Echo state network . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

60

11.1.5 Long short term memory network . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

60

11.1.6 Bi-directional RNN . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

60

11.1.7 Continuous-time RNN . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

60

11.1.8 Hierarchical RNN . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

60

11.1.9 Recurrent multilayer perceptron . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

61

11.1.10 Second Order Recurrent Neural Network . . . . . . . . . . . . . . . . . . . . . . . . . . .

61

11.1.11 Multiple Timescales Recurrent Neural Network (MTRNN) Model . . . . . . . . . . . . .

61

11.1.12 Pollack’s sequential cascaded networks . . . . . . . . . . . . . . . . . . . . . . . . . . .

61

11.1.13 Neural Turing Machines . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

61

11.1.14 Bidirectional Associative Memory (BAM) . . . . . . . . . . . . . . . . . . . . . . . . . .

61

11.2 Training . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

61

11.2.1 Gradient descent . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

61

11.2.2 Hessian Free Optimisation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

61

11.2.3 Global optimization methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

62

11.3 Related ﬁelds and models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

62

CONTENTS

v

11.4 Issues with recurrent neural networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

62

11.5 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

62

11.6 External links . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

64

12 Long short term memory

65

12.1 Architecture . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

65

12.2 Training . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

66

12.3 Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

66

12.4 See also . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

66

12.5 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

66

12.6 External links . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

67

13 Google Brain

68

13.1 History . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

68

13.2 In Google products . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

68

13.3 Team . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

68

13.4 Reception . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

68

13.5 See also . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

68

13.6 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

68

14 Google DeepMind

70

14.1 History . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

70

14.1.1 2011 to 2014 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

70

14.1.2 Acquisition by Google . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

70

14.2 Research . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

70

14.2.1 Deep reinforcement learning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

70

14.3 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

71

14.4 External links . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

71

15 Torch (machine learning)
15.1 torch

72

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

72

15.2 nn . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

72

15.3 Other packages . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

73

15.4 Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

73

15.5 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

73

15.6 External links . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

73

16 Theano (software)

74

16.1 See also . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

74

16.2 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

74

17 Deeplearning4j

75

17.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

75

17.2 Distributed . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

75

vi

CONTENTS
17.3 Scientiﬁc Computing for the JVM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

75

17.4 Canova Vectorization Lib for Machine-Learning . . . . . . . . . . . . . . . . . . . . . . . . . . .

75

17.5 Text & NLP . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

75

17.6 See also . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

75

17.7 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

76

17.8 External links . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

76

18 Gensim

77

18.1 Gensim's tagline . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

77

18.2 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

77

18.3 External links . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

77

19 Geoﬀrey Hinton

78

19.1 Career . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

78

19.2 Research interests . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

78

19.3 Honours and awards . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

78

19.4 Personal life . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

78

19.5 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

78

19.6 External links . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

79

20 Yann LeCun

80

20.1 Life . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

80

20.2 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

80

20.3 External links . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

81

21 Jürgen Schmidhuber

82

21.1 Contributions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

82

21.1.1 Recurrent neural networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

82

21.1.2 Artiﬁcial evolution / genetic programming . . . . . . . . . . . . . . . . . . . . . . . . . .

82

21.1.3 Neural economy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

82

21.1.4 Artiﬁcial curiosity and creativity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

82

21.1.5 Unsupervised learning / factorial codes . . . . . . . . . . . . . . . . . . . . . . . . . . . .

83

21.1.6 Kolmogorov complexity / computer-generated universe . . . . . . . . . . . . . . . . . . .

83

21.1.7 Universal AI . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

83

21.1.8 Low-complexity art / theory of beauty . . . . . . . . . . . . . . . . . . . . . . . . . . . .

83

21.1.9 Robot learning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

83

21.2 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

83

21.3 Sources . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

84

21.4 External links . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

84

22 Jeﬀ Dean (computer scientist)

85

22.1 Personal life and education . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

85

22.2 Career in computer science . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

85

CONTENTS

vii

22.3 Career at Google . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

85

22.4 Awards and honors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

85

22.5 Major publications

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

85

22.6 See also . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

85

22.7 External links . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

86

23 Andrew Ng
23.1 Machine learning research

87
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

87

23.2 Online education . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

87

23.3 Personal life . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

87

23.4 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

87

23.5 See also . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

88

23.6 External links . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

88

23.7 Text and image sources, contributors, and licenses . . . . . . . . . . . . . . . . . . . . . . . . . .

89

23.7.1 Text . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

89

23.7.2 Images . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

91

23.7.3 Content license . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

92

Chapter 1

Artiﬁcial neural network
“Neural network”redirects here. For networks of living neurons, see Biological neural network. For the journal, see Neural Networks (journal). For the evolutionary concept, see Neutral network (evolution).
In machine learning and cognitive science, artiﬁcial

tion is deﬁned by a set of input neurons which may be activated by the pixels of an input image. After being weighted and transformed by a function (determined by the network's designer), the activations of these neurons are then passed on to other neurons. This process is repeated until ﬁnally, an output neuron is activated. This determines which character was read.
Like other machine learning methods - systems that learn from data - neural networks have been used to solve a wide variety of tasks that are hard to solve using ordinary rule-based programming, including computer vision and speech recognition.

1.1 Background
Examinations of the human's central nervous system inspired the concept of neural networks. In an Artiﬁcial Neural Network, simple artiﬁcial nodes, known as
"neurons", neurodes” processing elements” “units”
“
“
,
or
, are connected together to form a network which mimics a biological neural network.
There is no single formal deﬁnition of what an artiﬁcial neural network is. However, a class of statistical models may commonly be called “Neural”if they possess the following characteristics:
An artiﬁcial neural network is an interconnected group of nodes, akin to the vast network of neurons in a brain. Here, each circular node represents an artiﬁcial neuron and an arrow represents a connection from the output of one neuron to the input of another.

1. consist of sets of adaptive weights, i.e. numerical parameters that are tuned by a learning algorithm, and 2. are capable of approximating non-linear functions of their inputs.

neural networks (ANNs) are a family of statistical learning models inspired by biological neural networks (the central nervous systems of animals, in particular the brain) and are used to estimate or approximate functions that can depend on a large number of inputs and are generally unknown. Artiﬁcial neural networks are generally presented as systems of interconnected "neurons" which send messages to each other. The connections have numeric weights that can be tuned based on experience, making neural nets adaptive to inputs and capable of learning.

The adaptive weights are conceptually connection strengths between neurons, which are activated during training and prediction.

Neural networks are similar to biological neural networks in performing functions collectively and in parallel by the units, rather than there being a clear delineation of subtasks to which various units are assigned. The term
“neural network”usually refers to models employed in
For example, a neural network for handwriting recogni- statistics, cognitive psychology and artiﬁcial intelligence.
1

2

CHAPTER 1. ARTIFICIAL NEURAL NETWORK

Neural network models which emulate the central ner- Neural network research stagnated after the publication vous system are part of theoretical neuroscience and of machine learning research by Marvin Minsky and computational neuroscience.
Seymour Papert* [7] (1969), who discovered two key isIn modern software implementations of artiﬁcial neu- sues with the computational machines that processed neural networks, the approach inspired by biology has been ral networks. The ﬁrst was that single-layer neural netlargely abandoned for a more practical approach based works were incapable of processing the exclusive-or ciron statistics and signal processing. In some of these sys- cuit. The second signiﬁcant issue was that computers tems, neural networks or parts of neural networks (like were not sophisticated enough to eﬀectively handle the long run time required by large neural networks. Neuartiﬁcial neurons) form components in larger systems that combine both adaptive and non-adaptive elements. While ral network research slowed until computers achieved greater processing power. Also key later advances was the more general approach of such systems is more suitable for real-world problem solving, it has little to do with the backpropagation algorithm which eﬀectively solved the exclusive-or problem (Werbos 1975).* [6] the traditional artiﬁcial intelligence connectionist models.
What they do have in common, however, is the princi- The parallel distributed processing of the mid-1980s beple of non-linear, distributed, parallel and local process- came popular under the name connectionism. The text by ing and adaptation. Historically, the use of neural net- David E. Rumelhart and James McClelland* [8] (1986) works models marked a paradigm shift in the late eight- provided a full exposition on the use of connectionism in ies from high-level (symbolic) AI, characterized by expert computers to simulate neural processes. systems with knowledge embodied in if-then rules, to low- Neural networks, as used in artiﬁcial intelligence, have level (sub-symbolic) machine learning, characterized by traditionally been viewed as simpliﬁed models of neural knowledge embodied in the parameters of a dynamical processing in the brain, even though the relation between system. this model and brain biological architecture is debated,

1.2 History

as it is not clear to what degree artiﬁcial neural networks mirror brain function.* [9]

Neural networks were gradually overtaken in popularity in machine learning by support vector machines and
*
Warren McCulloch and Walter Pitts [1] (1943) created other, much simpler methods such as linear classiﬁers. a computational model for neural networks based on Renewed interest in neural nets was sparked in the late mathematics and algorithms called threshold logic. This 2000s by the advent of deep learning. model paved the way for neural network research to split into two distinct approaches. One approach focused on biological processes in the brain and the other focused on the application of neural networks to artiﬁcial intelli- 1.2.1 Improvements since 2006 gence. Computational devices have been created in CMOS, for
In the late 1940s psychologist Donald Hebb* [2] created a both biophysical simulation and neuromorphic computhypothesis of learning based on the mechanism of neural ing. More recent eﬀorts show promise for creating plasticity that is now known as Hebbian learning. Heb- nanodevices* [10] for very large scale principal compobian learning is considered to be a 'typical' unsupervised nents analyses and convolution. If successful, these eflearning rule and its later variants were early models for forts could usher in a new era of neural computing* [11] long term potentiation. These ideas started being applied that is a step beyond digital computing, because it deto computational models in 1948 with Turing's B-type pends on learning rather than programming and because machines. it is fundamentally analog rather than digital even though
Farley and Wesley A. Clark* [3] (1954) ﬁrst used com- the ﬁrst instantiations may in fact be with CMOS digital putational machines, then called calculators, to simulate devices. a Hebbian network at MIT. Other neural network com- Between 2009 and 2012, the recurrent neural networks putational machines were created by Rochester, Holland, and deep feedforward neural networks developed in the
Habit, and Duda* [4] (1956). research group of Jürgen Schmidhuber at the Swiss AI
Frank Rosenblatt* [5] (1958) created the perceptron, an algorithm for pattern recognition based on a two-layer learning computer network using simple addition and subtraction. With mathematical notation, Rosenblatt also described circuitry not in the basic perceptron, such as the exclusive-or circuit, a circuit whose mathematical computation could not be processed until after the backpropagation algorithm was created by Paul Werbos* [6] (1975).

Lab IDSIA have won eight international competitions in pattern recognition and machine learning.* [12]* [13]
For example, the bi-directional and multi-dimensional long short term memory (LSTM)* [14]* [15]* [16]* [17] of
Alex Graves et al. won three competitions in connected handwriting recognition at the 2009 International Conference on Document Analysis and Recognition (ICDAR), without any prior knowledge about the three diﬀerent languages to be learned.

1.3. MODELS
Fast GPU-based implementations of this approach by
Dan Ciresan and colleagues at IDSIA have won several pattern recognition contests, including the IJCNN
2011 Traﬃc Sign Recognition Competition,* [18]* [19] the ISBI 2012 Segmentation of Neuronal Structures in
Electron Microscopy Stacks challenge,* [20] and others.
Their neural networks also were the ﬁrst artiﬁcial pattern recognizers to achieve human-competitive or even superhuman performance* [21] on important benchmarks such as traﬃc sign recognition (IJCNN 2012), or the MNIST handwritten digits problem of Yann LeCun at NYU.
Deep, highly nonlinear neural architectures similar to the
1980 neocognitron by Kunihiko Fukushima* [22] and the
“standard architecture of vision”,* [23] inspired by the simple and complex cells identiﬁed by David H. Hubel and Torsten Wiesel in the primary visual cortex, can also be pre-trained by unsupervised methods* [24]* [25] of Geoﬀ Hinton's lab at University of Toronto.* [26]* [27]
A team from this lab won a 2012 contest sponsored by
Merck to design software to help ﬁnd molecules that might lead to new drugs.* [28]

3
2. The learning process for updating the weights of the interconnections 3. The activation function that converts a neuron's weighted input to its output activation.
Mathematically, a neuron's network function f (x) is deﬁned as a composition of other functions gi (x) , which can further be deﬁned as a composition of other functions. This can be conveniently represented as a network structure, with arrows depicting the dependencies between variables. A widely used type of composition is the nonlinear weighted sum, where f (x) =
∑
K ( i wi gi (x)) , where K (commonly referred to as the activation function* [29]) is some predeﬁned function, such as the hyperbolic tangent. It will be convenient for the following to refer to a collection of functions gi as simply a vector g = (g1 , g2 , . . . , gn ) .

1.3 Models
Neural network models in artiﬁcial intelligence are usually referred to as artiﬁcial neural networks (ANNs); these are essentially simple mathematical models deﬁning a function f : X → Y or a distribution over X or both X and Y , but sometimes models are also intimately associated with a particular learning algorithm or learning rule. A common use of the phrase ANN model really means the deﬁnition of a class of such functions (where members of the class are obtained by varying parameters, connection weights, or speciﬁcs of the architecture such as the number of neurons or their connectivity).

1.3.1

Network function

See also: Graphical models
The word network in the term 'artiﬁcial neural network' refers to the inter–connections between the neurons in the diﬀerent layers of each system. An example system has three layers. The ﬁrst layer has input neurons which send data via synapses to the second layer of neurons, and then via more synapses to the third layer of output neurons. More complex systems will have more layers of neurons with some having increased layers of input neurons and output neurons. The synapses store parameters called “weights”that manipulate the data in the calculations.

ANN dependency graph

This ﬁgure depicts such a decomposition of f , with dependencies between variables indicated by arrows. These can be interpreted in two ways.
The ﬁrst view is the functional view: the input x is transformed into a 3-dimensional vector h , which is then transformed into a 2-dimensional vector g , which is ﬁnally transformed into f . This view is most commonly encountered in the context of optimization.
The second view is the probabilistic view: the random variable F = f (G) depends upon the random variable
G = g(H) , which depends upon H = h(X) , which depends upon the random variable X . This view is most commonly encountered in the context of graphical models.

The two views are largely equivalent. In either case, for this particular network architecture, the components of individual layers are independent of each other (e.g., the components of g are independent of each other given their input h ). This naturally enables a degree of parallelism
An ANN is typically deﬁned by three types of parameters: in the implementation.
Networks such as the previous one are commonly called
1. The interconnection pattern between the diﬀerent feedforward, because their graph is a directed acyclic layers of neurons graph. Networks with cycles are commonly called

4

CHAPTER 1. ARTIFICIAL NEURAL NETWORK
When N → ∞ some form of online machine learning must be used, where the cost is partially minimized as each new example is seen. While online machine learning is often used when D is ﬁxed, it is most useful in the case where the distribution changes slowly over time. In neural network methods, some form of online machine learning is frequently used for ﬁnite datasets.
See also: Mathematical optimization, Estimation theory and Machine learning

Choosing a cost function

Two separate depictions of the recurrent ANN dependency graph

While it is possible to deﬁne some arbitrary ad hoc cost function, frequently a particular cost will be used, either because it has desirable properties (such as convexity) or because it arises naturally from a particular formulation of the problem (e.g., in a probabilistic formulation the posterior probability of the model can be used as an inverse cost). Ultimately, the cost function will depend on the desired task. An overview of the three main categories of learning tasks is provided below:

recurrent. Such networks are commonly depicted in the manner shown at the top of the ﬁgure, where f is shown as being dependent upon itself. However, an implied tem- 1.3.3 poral dependence is not shown.

1.3.2

Learning

Learning paradigms

There are three major learning paradigms, each corresponding to a particular abstract learning task. These are supervised learning, unsupervised learning and reinforcement learning.

What has attracted the most interest in neural networks is the possibility of learning. Given a speciﬁc task to solve, and a class of functions F , learning means using a set Supervised learning of observations to ﬁnd f ∗ ∈ F which solves the task in some optimal sense.
In supervised learning, we are given a set of example pairs
This entails deﬁning a cost function C : F → R such that, (x, y), x ∈ X, y ∈ Y and the aim is to ﬁnd a function for the optimal solution f ∗ , C(f ∗ ) ≤ C(f ) ∀f ∈ F – f : X → Y in the allowed class of functions that matches
i.e., no solution has a cost less than the cost of the optimal the examples. In other words, we wish to infer the mapsolution (see Mathematical optimization). ping implied by the data; the cost function is related to the mismatch between our mapping and the data and it imThe cost function C is an important concept in learning, as it is a measure of how far away a particular solution plicitly contains prior knowledge about the problem domain. is from an optimal solution to the problem to be solved.
Learning algorithms search through the solution space to A commonly used cost is the mean-squared error, which tries to minimize the average squared error between the ﬁnd a function that has the smallest possible cost.
For applications where the solution is dependent on some network's output, f (x) , and the target value y over all data, the cost must necessarily be a function of the obser- the example pairs. When one tries to minimize this cost vations, otherwise we would not be modelling anything using gradient descent for the class of neural networks related to the data. It is frequently deﬁned as a statistic called multilayer perceptrons, one obtains the common to which only approximations can be made. As a sim- and well-known backpropagation algorithm for training ple example, consider the problem of ﬁnding the model neural networks.
[
] f , which minimizes C = E (f (x) − y)2 , for data Tasks that fall within the paradigm of supervised learnpairs (x, y) drawn from some distribution D . In prac- ing are pattern recognition (also known as classiﬁcation) tical situations we would only have N samples from D and regression (also known as function approximation). and thus, for the above example, we would only minimize The supervised learning paradigm is also applicable to
∑N
1
ˆ
C = N i=1 (f (xi )−yi )2 . Thus, the cost is minimized sequential data (e.g., for speech and gesture recognition). over a sample of the data rather than the entire data set. This can be thought of as learning with a “teacher”, in

1.4. EMPLOYING ARTIFICIAL NEURAL NETWORKS

5

the form of a function that provides continuous feedback those involved in vehicle routing,* [33] natural resources on the quality of solutions obtained thus far. management* [34]* [35] or medicine* [36] because of the ability of ANNs to mitigate losses of accuracy even when reducing the discretization grid density for numerically
Unsupervised learning approximating the solution of the original control problems.
In unsupervised learning, some data x is given and the cost function to be minimized, that can be any function Tasks that fall within the paradigm of reinforcement learning are control problems, games and other sequential of the data x and the network's output, f . decision making tasks.
The cost function is dependent on the task (what we are trying to model) and our a priori assumptions (the implicit See also: dynamic programming and stochastic control properties of our model, its parameters and the observed variables). As a trivial example, consider the model f (x) = a where a is a constant and the cost C = E[(x − f (x))2 ] . Minimizing this cost will give us a value of a that is equal to the mean of the data. The cost function can be much more complicated. Its form depends on the application: for example, in compression it could be related to the mutual information between x and f (x) , whereas in statistical modeling, it could be related to the posterior probability of the model given the data (note that in both of those examples those quantities would be maximized rather than minimized). Tasks that fall within the paradigm of unsupervised learning are in general estimation problems; the applications include clustering, the estimation of statistical distributions, compression and ﬁltering.

1.3.4 Learning algorithms
Training a neural network model essentially means selecting one model from the set of allowed models (or, in a Bayesian framework, determining a distribution over the set of allowed models) that minimizes the cost criterion. There are numerous algorithms available for training neural network models; most of them can be viewed as a straightforward application of optimization theory and statistical estimation.
Most of the algorithms used in training artiﬁcial neural networks employ some form of gradient descent, using backpropagation to compute the actual gradients. This is done by simply taking the derivative of the cost function with respect to the network parameters and then changing those parameters in a gradient-related direction.

Evolutionary methods,* [37] gene expression programming,* [38] simulated annealing,* [39] expectationIn reinforcement learning, data x are usually not given, maximization, non-parametric methods and particle
*
but generated by an agent's interactions with the environ- swarm optimization [40] are some commonly used ment. At each point in time t , the agent performs an methods for training neural networks. action yt and the environment generates an observation See also: machine learning xt and an instantaneous cost ct , according to some (usually unknown) dynamics. The aim is to discover a policy for selecting actions that minimizes some measure of a long-term cost; i.e., the expected cumulative cost. The 1.4 Employing artiﬁcial neural netenvironment's dynamics and the long-term cost for each works policy are usually unknown, but can be estimated.
Reinforcement learning

More formally the environment is modelled as a Markov decision process (MDP) with states s1 , ..., sn ∈ S and actions a1 , ..., am ∈ A with the following probability distributions: the instantaneous cost distribution P (ct |st ) , the observation distribution P (xt |st ) and the transition
P (st+1 |st , at ) , while a policy is deﬁned as conditional distribution over actions given the observations. Taken together, the two then deﬁne a Markov chain (MC). The aim is to discover the policy that minimizes the cost; i.e., the MC for which the cost is minimal.
ANNs are frequently used in reinforcement learning as part of the overall algorithm.* [30]* [31] Dynamic programming has been coupled with ANNs (Neuro dynamic programming) by Bertsekas and Tsitsiklis* [32] and applied to multi-dimensional nonlinear problems such as

Perhaps the greatest advantage of ANNs is their ability to be used as an arbitrary function approximation mechanism that 'learns' from observed data. However, using them is not so straightforward, and a relatively good understanding of the underlying theory is essential.
• Choice of model: This will depend on the data representation and the application. Overly complex models tend to lead to problems with learning.
• Learning algorithm: There are numerous trade-oﬀs between learning algorithms. Almost any algorithm will work well with the correct hyperparameters for training on a particular ﬁxed data set. However, selecting and tuning an algorithm for training on un-

6

CHAPTER 1. ARTIFICIAL NEURAL NETWORK

seen data requires a signiﬁcant amount of experi- These networks have also been used to diagnose prostate mentation. cancer. The diagnoses can be used to make speciﬁc models taken from a large group of patients compared to in• Robustness: If the model, cost function and learn- formation of one given patient. The models do not deing algorithm are selected appropriately the result- pend on assumptions about correlations of diﬀerent variing ANN can be extremely robust. ables. Colorectal cancer has also been predicted using the neural networks. Neural networks could predict the
With the correct implementation, ANNs can be used nat- outcome for a patient with colorectal cancer with more urally in online learning and large data set applications. accuracy than the current clinical methods. After trainTheir simple implementation and the existence of mostly ing, the networks could predict multiple patient outcomes local dependencies exhibited in the structure allows for from unrelated institutions.* [43] fast, parallel implementations in hardware.

1.5 Applications

1.5.2 Neural networks and neuroscience

The utility of artiﬁcial neural network models lies in the fact that they can be used to infer a function from observations. This is particularly useful in applications where the complexity of the data or task makes the design of such a function by hand impractical.

Theoretical and computational neuroscience is the ﬁeld concerned with the theoretical analysis and the computational modeling of biological neural systems. Since neural systems are intimately related to cognitive processes and behavior, the ﬁeld is closely related to cognitive and behavioral modeling.

The aim of the ﬁeld is to create models of biological neural systems in order to understand how biological systems work. To gain this understanding, neuroscientists strive
The tasks artiﬁcial neural networks are applied to tend to to make a link between observed biological processes fall within the following broad categories:
(data), biologically plausible mechanisms for neural processing and learning (biological neural network models)
• Function approximation, or regression analysis, in- and theory (statistical learning theory and information cluding time series prediction, ﬁtness approximation theory). and modeling.

1.5.1

Real-life applications

• Classiﬁcation, including pattern and sequence recognition, novelty detection and sequential deci- Types of models sion making.
• Data processing, including ﬁltering, clustering, blind Many models are used in the ﬁeld, deﬁned at diﬀerent levels of abstraction and modeling diﬀerent aspects of neural source separation and compression. systems. They range from models of the short-term be• Robotics, including directing manipulators, havior of individual neurons, models of how the dynamics prosthesis. of neural circuitry arise from interactions between individual neurons and ﬁnally to models of how behavior can
• Control, including Computer numerical control. arise from abstract neural modules that represent complete subsystems. These include models of the long-term,
Application areas include the system identiﬁcation and and short-term plasticity, of neural systems and their relacontrol (vehicle control, process control, natural re- tions to learning and memory from the individual neuron sources management), quantum chemistry,* [41] game- to the system level. playing and decision making (backgammon, chess, poker), pattern recognition (radar systems, face identiﬁcation, object recognition and more), sequence recognition (gesture, speech, handwritten text recognition),
1.6 Neural network software medical diagnosis, ﬁnancial applications (e.g. automated trading systems), data mining (or knowledge discovery in databases, “KDD”), visualization and e-mail spam ﬁl- Main article: Neural network software tering. Artiﬁcial neural networks have also been used to diagnose several cancers. An ANN based hybrid lung cancer detection system named HLND improves the accuracy of diagnosis and the speed of lung cancer radiology.* [42]

Neural network software is used to simulate, research, develop and apply artiﬁcial neural networks, biological neural networks and, in some cases, a wider array of adaptive systems.

1.8. THEORETICAL PROPERTIES

7

1.7 Types of artiﬁcial neural networks

regarding convergence are an unreliable guide to practical application. Main article: Types of artiﬁcial neural networks

1.8.4 Generalization and statistics

Artiﬁcial neural network types vary from those with only one or two layers of single direction logic, to complicated multi–input many directional feedback loops and layers.
On the whole, these systems use algorithms in their programming to determine control and organization of their functions. Most systems use “weights”to change the parameters of the throughput and the varying connections to the neurons. Artiﬁcial neural networks can be autonomous and learn by input from outside “teachers” or even self-teaching from written-in rules.

In applications where the goal is to create a system that generalizes well in unseen examples, the problem of overtraining has emerged. This arises in convoluted or overspeciﬁed systems when the capacity of the network signiﬁcantly exceeds the needed free parameters. There are two schools of thought for avoiding this problem:
The ﬁrst is to use cross-validation and similar techniques to check for the presence of overtraining and optimally select hyperparameters such as to minimize the generalization error. The second is to use some form of regularization. This is a concept that emerges naturally in a probabilistic (Bayesian) framework, where the regularization can be performed by selecting a larger prior prob1.8 Theoretical properties ability over simpler models; but also in statistical learning theory, where the goal is to minimize over two quantities: the 'empirical risk' and the 'structural risk', which roughly
1.8.1 Computational power corresponds to the error over the training set and the preThe multi-layer perceptron (MLP) is a universal function dicted error in unseen data due to overﬁtting. approximator, as proven by the universal approximation theorem. However, the proof is not constructive regarding the number of neurons required or the settings of the weights. Work by Hava Siegelmann and Eduardo D. Sontag has provided a proof that a speciﬁc recurrent architecture with rational valued weights (as opposed to full precision real number-valued weights) has the full power of a
Universal Turing Machine* [44] using a ﬁnite number of neurons and standard linear connections. Further, it has been shown that the use of irrational values for weights results in a machine with super-Turing power.* [45]

1.8.2

Capacity

Artiﬁcial neural network models have a property called
'capacity', which roughly corresponds to their ability to model any given function. It is related to the amount of information that can be stored in the network and to the notion of complexity.

Conﬁdence analysis of a neural network

Supervised neural networks that use a mean squared error
(MSE) cost function can use formal statistical methods to determine the conﬁdence of the trained model. The MSE on a validation set can be used as an estimate for variance.
This value can then be used to calculate the conﬁdence interval of the output of the network, assuming a normal distribution. A conﬁdence analysis made this way is sta1.8.3 Convergence tistically valid as long as the output probability distribuNothing can be said in general about convergence since it tion stays the same and the network is not modiﬁed. depends on a number of factors. Firstly, there may exist By assigning a softmax activation function, a generalizamany local minima. This depends on the cost function tion of the logistic function, on the output layer of the and the model. Secondly, the optimization method used neural network (or a softmax component in a componentmight not be guaranteed to converge when far away from based neural network) for categorical target variables, the a local minimum. Thirdly, for a very large amount of outputs can be interpreted as posterior probabilities. This data or parameters, some methods become impractical. is very useful in classiﬁcation as it gives a certainty meaIn general, it has been found that theoretical guarantees sure on classiﬁcations.

8

CHAPTER 1. ARTIFICIAL NEURAL NETWORK

The softmax activation function is:

(they tend to consume considerable amounts of time and money). ex i yi = ∑c

Computing power continues to grow roughly according to Moore's Law, which may provide suﬃcient resources to accomplish new tasks. Neuromorphic engineering addresses the hardware diﬃculty directly, by constructing non-Von-Neumann chips with circuits designed to implement neural nets from the ground up.

j=1

ex j

1.9 Controversies
1.9.1

Training issues

A common criticism of neural networks, particularly in robotics, is that they require a large diversity of training for real-world operation . This is not surprising, since any learning machine needs suﬃcient representative examples in order to capture the underlying structure that allows it to generalize to new cases. Dean Pomerleau, in his research presented in the paper “Knowledge-based
Training of Artiﬁcial Neural Networks for Autonomous
Robot Driving,”uses a neural network to train a robotic vehicle to drive on multiple types of roads (single lane, multi-lane, dirt, etc.). A large amount of his research is devoted to (1) extrapolating multiple training scenarios from a single training experience, and (2) preserving past training diversity so that the system does not become overtrained (if, for example, it is presented with a series of right turns – it should not learn to always turn right).
These issues are common in neural networks that must decide from amongst a wide variety of responses, but can be dealt with in several ways, for example by randomly shufﬂing the training examples, by using a numerical optimization algorithm that does not take too large steps when changing the network connections following an example, or by grouping examples in so-called mini-batches.

1.9.3 Practical counterexamples to criticisms
Arguments against Dewdney's position are that neural networks have been successfully used to solve many complex and diverse tasks, ranging from autonomously ﬂying aircraft* [46] to detecting credit card fraud .
Technology writer Roger Bridgman commented on
Dewdney's statements about neural nets:
Neural networks, for instance, are in the dock not only because they have been hyped to high heaven, (what hasn't?) but also because you could create a successful net without understanding how it worked: the bunch of numbers that captures its behaviour would in all probability be “an opaque, unreadable table...valueless as a scientiﬁc resource”.
In spite of his emphatic declaration that science is not technology, Dewdney seems here to pillory neural nets as bad science when most of those devising them are just trying to be good engineers. An unreadable table that a useful machine could read would still be well worth having.* [47]

A. K. Dewdney, a former Scientiﬁc American columnist, wrote in 1997, “Although neural nets do solve a few toy problems, their powers of computation are so limited that
I am surprised anyone takes them seriously as a general problem-solving tool.”(Dewdney, p. 82)
Although it is true that analyzing what has been learned by an artiﬁcial neural network is diﬃcult, it is much easier to do so than to analyze what has been learned by a
1.9.2 Hardware issues biological neural network. Furthermore, researchers involved in exploring learning algorithms for neural netTo implement large and eﬀective software neural networks are gradually uncovering generic principles which works, considerable processing and storage resources allow a learning machine to be successful. For example, need to be committed . While the brain has hardware
Bengio and LeCun (2007) wrote an article regarding local tailored to the task of processing signals through a graph vs non-local learning, as well as shallow vs deep architecof neurons, simulating even a most simpliﬁed form on ture.* [48]
Von Neumann technology may compel a neural network designer to ﬁll many millions of database rows for its connections – which can consume vast amounts of computer memory and hard disk space. Furthermore, the designer 1.9.4 Hybrid approaches of neural network systems will often need to simulate the transmission of signals through many of these con- Some other criticisms come from advocates of hybrid nections and their associated neurons – which must often models (combining neural networks and symbolic apbe matched with incredible amounts of CPU processing proaches), who believe that the intermix of these two appower and time. While neural networks often yield eﬀec- proaches can better capture the mechanisms of the human tive programs, they too often do so at the cost of eﬃciency mind.* [49]* [50]

1.12. REFERENCES

1.10 Gallery
• A single-layer feedforward artiﬁcial neural network.
Arrows originating from are omitted for clarity.
There are p inputs to this network and q outputs.
In this system, the value of the qth output, would be calculated as
• A two-layer feedforward artiﬁcial neural network.
•
•

1.11 See also

9
• Habituation
• In Situ Adaptive Tabulation
• Models of neural computation
• Multilinear subspace learning
• Neuroevolution
• Neural coding
• Neural gas
• Neural network software
• Neuroscience
• Ni1000 chip

• 20Q

• Nonlinear system identiﬁcation

• ADALINE

• Optical neural network

• Adaptive resonance theory

• Parallel Constraint Satisfaction Processes

• Artiﬁcial life

• Parallel distributed processing

• Associative memory

• Radial basis function network

• Autoencoder

• Recurrent neural networks

• Backpropagation

• Self-organizing map

• BEAM robotics

• Spiking neural network

• Biological cybernetics

• Systolic array

• Biologically inspired computing

• Tensor product network

• Blue brain

• Time delay neural network (TDNN)

• Catastrophic interference
• Cerebellar Model Articulation Controller
• Cognitive architecture
• Cognitive science
• Convolutional neural network (CNN)
• Connectionist expert system
• Connectomics
• Cultured neuronal networks
• Deep learning
• Digital morphogenesis
• Encog
• Fuzzy logic
• Gene expression programming
• Genetic algorithm
• Group method of data handling

1.12 References
[1] McCulloch, Warren; Walter Pitts (1943). “A Logical Calculus of Ideas Immanent in Nervous Activity”
. Bulletin of Mathematical Biophysics 5 (4): 115–133. doi:10.1007/BF02478259. [2] Hebb, Donald (1949). The Organization of Behavior.
New York: Wiley.
[3] Farley, B.G.; W.A. Clark (1954). “Simulation of
Self-Organizing Systems by Digital Computer”. IRE
Transactions on Information Theory 4 (4): 76–84. doi:10.1109/TIT.1954.1057468. [4] Rochester, N.; J.H. Holland, L.H. Habit, and W.L.
Duda (1956). “Tests on a cell assembly theory of the action of the brain, using a large digital computer”.
IRE Transactions on Information Theory 2 (3): 80–93. doi:10.1109/TIT.1956.1056810. [5] Rosenblatt, F. (1958). “The Perceptron: A Probabilistic Model For Information Storage And Organization In
The Brain”. Psychological Review 65 (6): 386–408. doi:10.1037/h0042519. PMID 13602029.

10

[6] Werbos, P.J. (1975). Beyond Regression: New Tools for
Prediction and Analysis in the Behavioral Sciences.
[7] Minsky, M.; S. Papert (1969). An Introduction to Computational Geometry. MIT Press. ISBN 0-262-63022-2.
[8] Rumelhart, D.E; James McClelland (1986). Parallel Distributed Processing: Explorations in the Microstructure of
Cognition. Cambridge: MIT Press.
[9] Russell, Ingrid. “Neural Networks Module”. Retrieved
2012.

CHAPTER 1. ARTIFICIAL NEURAL NETWORK

[20] D. Ciresan, A. Giusti, L. Gambardella, J. Schmidhuber.
Deep Neural Networks Segment Neuronal Membranes in
Electron Microscopy Images. In Advances in Neural Information Processing Systems (NIPS 2012), Lake Tahoe,
2012.
[21] D. C. Ciresan, U. Meier, J. Schmidhuber. Multi-column
Deep Neural Networks for Image Classiﬁcation. IEEE
Conf. on Computer Vision and Pattern Recognition
CVPR 2012.

[10] Yang, J. J.; Pickett, M. D.; Li, X. M.; Ohlberg, D. A. A.;
Stewart, D. R.; Williams, R. S. Nat. Nanotechnol. 2008,
3, 429–433.

[22] Fukushima, K. (1980).“Neocognitron: A self-organizing neural network model for a mechanism of pattern recognition unaﬀected by shift in position”. Biological Cybernetics 36 (4): 93–202. doi:10.1007/BF00344251. PMID
7370364.

[11] Strukov, D. B.; Snider, G. S.; Stewart, D. R.; Williams, R.
S. Nature 2008, 453, 80–83.

[23] M Riesenhuber, T Poggio. Hierarchical models of object recognition in cortex. Nature neuroscience, 1999.

[12] 2012 Kurzweil AI Interview with Jürgen Schmidhuber on the eight competitions won by his Deep Learning team
2009–2012

[24] Deep belief networks at Scholarpedia.

[14] Graves, Alex; and Schmidhuber, Jürgen; Oﬄine Handwriting Recognition with Multidimensional Recurrent Neural Networks, in Bengio, Yoshua; Schuurmans, Dale; Lafferty, John; Williams, Chris K. I.; and Culotta, Aron
(eds.), Advances in Neural Information Processing Systems
22 (NIPS'22), 7–10 December 2009, Vancouver, BC, Neural Information Processing Systems (NIPS) Foundation,
2009, pp. 545–552.

[27] Hinton, G. E.; Osindero, S.; Teh, Y. (2006).
“A fast learning algorithm for deep belief nets”
(PDF). Neural Computation 18 (7):
1527–1554.
doi:10.1162/neco.2006.18.7.1527. PMID 16764513.

[15] A. Graves, M. Liwicki, S. Fernandez, R. Bertolami, H.
Bunke, J. Schmidhuber. A Novel Connectionist System for Improved Unconstrained Handwriting Recognition.
IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 31, no. 5, 2009.

[30] Dominic, S., Das, R., Whitley, D., Anderson, C. (July
1991). “Genetic reinforcement learning for neural networks”. IJCNN-91-Seattle International Joint Conference on Neural Networks. IJCNN-91-Seattle International
Joint Conference on Neural Networks. Seattle, Washington, USA: IEEE. doi:10.1109/IJCNN.1991.155315.
ISBN 0-7803-0164-1. Retrieved 29 July 2012.

[25] Hinton, G. E.; Osindero, S.; Teh, Y. W. (2006).
“A Fast Learning Algorithm for Deep Belief Nets”
[13] http://www.kurzweilai.net/
(PDF). Neural Computation 18 (7): 1527–1554. how-bio-inspired-deep-learning-keeps-winning-competitions doi:10.1162/neco.2006.18.7.1527. PMID 16764513.
2012 Kurzweil AI Interview with Jürgen Schmidhuber on the eight competitions won by his Deep Learning team [26] http://www.scholarpedia.org/article/Deep_belief_ networks /
2009–2012

[16] Graves, Alex; and Schmidhuber, Jürgen; Oﬄine Handwriting Recognition with Multidimensional Recurrent Neural Networks, in Bengio, Yoshua; Schuurmans, Dale; Lafferty, John; Williams, Chris K. I.; and Culotta, Aron
(eds.), Advances in Neural Information Processing Systems
22 (NIPS'22), December 7th–10th, 2009, Vancouver, BC,
Neural Information Processing Systems (NIPS) Foundation, 2009, pp. 545–552
[17] A. Graves, M. Liwicki, S. Fernandez, R. Bertolami, H.
Bunke, J. Schmidhuber. A Novel Connectionist System for Improved Unconstrained Handwriting Recognition.
IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 31, no. 5, 2009.
[18] D. C. Ciresan, U. Meier, J. Masci, J. Schmidhuber. MultiColumn Deep Neural Network for Traﬃc Sign Classiﬁcation. Neural Networks, 2012.
[19] D. C. Ciresan, U. Meier, J. Masci, J. Schmidhuber. MultiColumn Deep Neural Network for Traﬃc Sign Classiﬁcation. Neural Networks, 2012.

[28] John Markoﬀ (November 23, 2012). “Scientists See
Promise in Deep-Learning Programs”. New York Times.
[29] “The Machine Learning Dictionary”.

[31] Hoskins, J.C.; Himmelblau, D.M. (1992). “Process control via artiﬁcial neural networks and reinforcement learning”. Computers & Chemical Engineering 16 (4): 241–
251. doi:10.1016/0098-1354(92)80045-B.
[32] Bertsekas, D.P., Tsitsiklis, J.N. (1996). Neuro-dynamic programming. Athena Scientiﬁc. p. 512. ISBN 1886529-10-8.
[33] Secomandi, Nicola (2000). “Comparing neuro-dynamic programming algorithms for the vehicle routing problem with stochastic demands”. Computers & Operations
Research 27 (11–12): 1201–1225. doi:10.1016/S03050548(99)00146-X.
[34] de Rigo, D., Rizzoli, A. E., Soncini-Sessa, R., Weber, E.,
Zenesi, P. (2001). “Neuro-dynamic programming for the eﬃcient management of reservoir networks” (PDF).
Proceedings of MODSIM 2001, International Congress on
Modelling and Simulation. MODSIM 2001, International

1.13. BIBLIOGRAPHY

Congress on Modelling and Simulation. Canberra, Australia: Modelling and Simulation Society of Australia and New Zealand. doi:10.5281/zenodo.7481. ISBN 0867405252. Retrieved 29 July 2012.
[35] Damas, M., Salmeron, M., Diaz, A., Ortega, J., Prieto, A., Olivares, G. (2000). “Genetic algorithms and neuro-dynamic programming: application to water supply networks”. Proceedings of 2000 Congress on Evolutionary Computation. 2000 Congress on Evolutionary Computation. La Jolla, California, USA: IEEE. doi:10.1109/CEC.2000.870269. ISBN 0-7803-6375-2.
Retrieved 29 July 2012.
[36] Deng, Geng; Ferris, M.C. (2008). “Neuro-dynamic programming for fractionated radiotherapy planning”.
Springer Optimization and Its Applications 12: 47–70. doi:10.1007/978-0-387-73299-2_3. [37] de Rigo, D., Castelletti, A., Rizzoli, A.E., Soncini-Sessa,
R., Weber, E. (January 2005). “A selective improvement technique for fastening Neuro-Dynamic Programming in Water Resources Network Management”. In
Pavel Zítek. Proceedings of the 16th IFAC World Congress
– IFAC-PapersOnLine. 16th IFAC World Congress 16.
Prague, Czech Republic: IFAC. doi:10.3182/200507036-CZ-1902.02172. ISBN 978-3-902661-75-3. Retrieved
30 December 2011.
[38] Ferreira, C. (2006). “Designing Neural Networks Using
Gene Expression Programming”(PDF). In A. Abraham,
B. de Baets, M. Köppen, and B. Nickolay, eds., Applied
Soft Computing Technologies: The Challenge of Complexity, pages 517–536, Springer-Verlag.
[39] Da, Y., Xiurun, G. (July 2005). T. Villmann, ed. An improved PSO-based ANN with simulated annealing technique. New Aspects in Neurocomputing: 11th European Symposium on Artiﬁcial Neural Networks. Elsevier. doi:10.1016/j.neucom.2004.07.002. [40] Wu, J., Chen, E. (May 2009). Wang, H., Shen, Y., Huang,
T., Zeng, Z., ed. A Novel Nonparametric Regression Ensemble for Rainfall Forecasting Using Particle Swarm Optimization Technique Coupled with Artiﬁcial Neural Network. 6th International Symposium on Neural Networks,
ISNN 2009. Springer. doi:10.1007/978-3-642-015137_6. ISBN 978-3-642-01215-0.
[41] Roman M. Balabin, Ekaterina I. Lomakina (2009).“Neural network approach to quantum-chemistry data: Accurate prediction of density functional theory energies”. J.
Chem. Phys. 131 (7): 074104. doi:10.1063/1.3206326.
PMID 19708729.
[42] Ganesan, N. “Application of Neural Networks in Diagnosing Cancer Disease Using Demographic Data”(PDF).
International Journal of Computer Applications.
[43] Bottaci, Leonardo. “Artiﬁcial Neural Networks Applied to Outcome Prediction for Colorectal Cancer Patients in
Separate Institutions” (PDF). The Lancet.
[44] Siegelmann, H.T.; Sontag, E.D. (1991). “Turing computability with neural nets” (PDF). Appl. Math. Lett. 4
(6): 77–80. doi:10.1016/0893-9659(91)90080-F.

11

[45] Balcázar, José (Jul 1997). “Computational Power of
Neural Networks: A Kolmogorov Complexity Characterization”. Information Theory, IEEE Transactions on 43
(4): 1175–1183. doi:10.1109/18.605580. Retrieved 3
November 2014.
[46] NASA - Dryden Flight Research Center - News
Room: News Releases: NASA NEURAL NETWORK
PROJECT PASSES MILESTONE. Nasa.gov. Retrieved on 2013-11-20.
[47] Roger Bridgman's defence of neural networks
[48] http://www.iro.umontreal.ca/~{}lisa/publications2/ index.php/publications/show/4 [49] Sun and Bookman (1990)
[50] Tahmasebi; Hezarkhani (2012). “A hybrid neural networks-fuzzy logic-genetic algorithm for grade estimation”. Computers & Geosciences 42: 18–27. doi:10.1016/j.cageo.2012.02.004. 1.13 Bibliography
• Bhadeshia H. K. D. H. (1999).
“Neural Networks in Materials Science” (PDF).
ISIJ International 39 (10):
966–979.
doi:10.2355/isijinternational.39.966.
• Bishop, C.M. (1995) Neural Networks for Pattern Recognition, Oxford: Oxford University Press.
ISBN 0-19-853849-9 (hardback) or ISBN 0-19853864-2 (paperback)
• Cybenko, G.V. (1989). Approximation by Superpositions of a Sigmoidal function, Mathematics of
Control, Signals, and Systems, Vol. 2 pp. 303–314. electronic version
• Duda, R.O., Hart, P.E., Stork, D.G. (2001) Pattern classiﬁcation (2nd edition), Wiley, ISBN 0-47105669-3
• Egmont-Petersen, M., de Ridder, D., Handels, H.
(2002). “Image processing with neural networks – a review” Pattern Recognition 35 (10): 2279–2301.
.
doi:10.1016/S0031-3203(01)00178-9.
• Gurney, K. (1997) An Introduction to Neural Networks London: Routledge. ISBN 1-85728-673-1
(hardback) or ISBN 1-85728-503-4 (paperback)
• Haykin, S. (1999) Neural Networks: A Comprehensive Foundation, Prentice Hall, ISBN 0-13-2733501
• Fahlman, S, Lebiere, C (1991). The CascadeCorrelation Learning Architecture, created for
National Science Foundation, Contract Number
EET-8716324, and Defense Advanced Research
Projects Agency (DOD), ARPA Order No. 4976 under Contract F33615-87-C-1499. electronic version 12
• Hertz, J., Palmer, R.G., Krogh. A.S. (1990) Introduction to the theory of neural computation, Perseus
Books. ISBN 0-201-51560-1
• Lawrence, Jeanette (1994) Introduction to Neural Networks, California Scientiﬁc Software Press.
ISBN 1-883157-00-5
• Masters, Timothy (1994) Signal and Image Processing with Neural Networks, John Wiley & Sons, Inc.
ISBN 0-471-04963-8
• Ripley, Brian D. (1996) Pattern Recognition and
Neural Networks, Cambridge
• Siegelmann, H.T. and Sontag, E.D. (1994). Analog computation via neural networks, Theoretical
Computer Science, v. 131, no. 2, pp. 331–360. electronic version
• Sergios Theodoridis, Konstantinos Koutroumbas
(2009) “Pattern Recognition”, 4th Edition, Academic Press, ISBN 978-1-59749-272-0.
• Smith, Murray (1993) Neural Networks for Statistical Modeling, Van Nostrand Reinhold, ISBN 0-44201310-8
• Wasserman, Philip (1993) Advanced Methods in
Neural Computing, Van Nostrand Reinhold, ISBN
0-442-00461-3
• Computational Intelligence:
A Methodological Introduction by Kruse, Borgelt, Klawonn,
Moewes, Steinbrecher, Held, 2013, Springer, ISBN
9781447150121
• Neuro-Fuzzy-Systeme (3rd edition) by Borgelt,
Klawonn, Kruse, Nauck, 2003, Vieweg, ISBN
9783528252656

1.14 External links
• Neural Networks at DMOZ
• A brief introduction to Neural Networks (PDF), illustrated 250p textbook covering the common kinds of neural networks (CC license).

CHAPTER 1. ARTIFICIAL NEURAL NETWORK

Chapter 2

Deep learning
For deep versus shallow learning in educational psychology, see Student approaches to learning

2.1 Introduction
2.1.1 Deﬁnitions

Deep learning (deep machine learning, or deep structured learning, or hierarchical learning, or sometimes DL) is a branch of machine learning based on a set of algorithms that attempt to model high-level abstractions in data by using model architectures, with complex structures or otherwise, composed of multiple non-linear transformations.* [1]* (p198)* [2]* [3]* [4]

There are a number of ways that the ﬁeld of deep learning has been characterized. Deep learning is a class of machine learning training algorithms that* [1]* (pp199–
200)
• use a cascade of many layers of nonlinear processing units for feature extraction and transformation.
The next layer uses the output from the previous layer as input. The algorithms may be supervised or unsupervised and applications include pattern recognition and statistical classiﬁcation.

Deep learning is part of a broader family of machine learning methods based on learning representations of data. An observation (e.g., an image) can be represented in many ways such as a vector of intensity values per pixel, or in a more abstract way as a set of edges, regions of particular shape, etc. Some representations make it easier to learn tasks (e.g., face recognition or facial expression recognition* [5]) from examples. One of the promises of deep learning is replacing handcrafted features with eﬃcient algorithms for unsupervised or semi-supervised feature learning and hierarchical feature extraction.* [6]
Research in this area attempts to make better representations and create models to learn these representations from large-scale unlabeled data. Some of the representations are inspired by advances in neuroscience and are loosely based on interpretation of information processing and communication patterns in a nervous system, such as neural coding which attempts to deﬁne a relationship between the stimulus and the neuronal responses and the relationship among the electrical activity of the neurons in the brain.* [7]
Various deep learning architectures such as deep neural networks, convolutional deep neural networks, and deep belief networks have been applied to ﬁelds like computer vision, automatic speech recognition, natural language processing, audio recognition and bioinformatics where they have been shown to produce state-of-the-art results on various tasks.
Alternatively, deep learning has been characterized as a buzzword, or a rebranding of neural networks.* [8]* [9]

• are based on the (unsupervised) learning of multiple levels of features or representations of the data.
Higher level features are derived from lower level features to form a hierarchical representation.
• are part of the broader machine learning ﬁeld of learning representations of data.
• learn multiple levels of representations that correspond to diﬀerent levels of abstraction; the levels form a hierarchy of concepts.
These deﬁnitions have in common (1) multiple layers of nonlinear processing units and (2) the supervised or unsupervised learning of feature representations in each layer, with the layers forming a hierarchy from low-level to high-level features.* [1]* (p200) The composition of a layer of nonlinear processing units used in a deep belief algorithm depends on the problem to be solved. Layers that have been used in deep learning include hidden layers of an artiﬁcial neural network, restricted Boltzmann machines, and sets of complicated propositional formulas.* [2]
Deep learning algorithms are contrasted with shallow learning algorithms by the number of parameterized transformations a signal encounters as it propagates from the input layer to the output layer, where a parameterized transformation is a processing unit that has trainable parameters, such as weights and thresholds.* [4]* (p6) A

13

14

CHAPTER 2. DEEP LEARNING

chain of transformations from input to output is a credit assignment path (CAP). CAPs describe potentially causal connections between input and output and may vary in length. For a feedforward neural network, the depth of the CAPs, and thus the depth of the network, is the number of hidden layers plus one (the output layer is also parameterized). For recurrent neural networks, in which a signal may propagate through a layer more than once, the
CAP is potentially unlimited in length. There is no universally agreed upon threshold of depth dividing shallow learning from deep learning, but most researchers in the ﬁeld agree that deep learning has multiple nonlinear layers (CAP > 2) and Schmidhuber considers CAP > 10 to be very deep learning.* [4]* (p7)

2.1.2

Fundamental concepts

rithms can make use of the unlabeled data that supervised algorithms cannot. Unlabeled data is usually more abundant than labeled data, making this an important beneﬁt of these algorithms. The deep belief network is an example of a deep structure that can be trained in an unsupervised manner.* [3]

2.2 History
Deep learning architectures, speciﬁcally those built from artiﬁcial neural networks (ANN), date back at least to the Neocognitron introduced by Kunihiko Fukushima in
1980.* [10] The ANNs themselves date back even further.
In 1989, Yann LeCun et al. were able to apply the standard backpropagation algorithm, which had been around since 1974,* [11] to a deep neural network with the purpose of recognizing handwritten ZIP codes on mail. Despite the success of applying the algorithm, the time to train the network on this dataset was approximately 3 days, making it impractical for general use.* [12] Many factors contribute to the slow speed, one being due to the so-called vanishing gradient problem analyzed in 1991 by
Sepp Hochreiter.* [13]* [14]

Deep learning algorithms are based on distributed representations. The underlying assumption behind distributed representations is that observed data is generated by the interactions of many diﬀerent factors on diﬀerent levels. Deep learning adds the assumption that these factors are organized into multiple levels, corresponding to diﬀerent levels of abstraction or composition. Varying numbers of layers and layer sizes can be used to provide While such neural networks by 1991 were used for recognizing isolated 2-D hand-written digits, 3-D object diﬀerent amounts of abstraction.* [3] recognition by 1991 used a 3-D model-based approach
Deep learning algorithms in particular exploit this idea – matching 2-D images with a handcrafted 3-D object of hierarchical explanatory factors. Diﬀerent concepts model. Juyang Weng et al.. proposed that a human brain are learned from other concepts, with the more abstract, does not use a monolithic 3-D object model and in 1992 higher level concepts being learned from the lower level they published Cresceptron,* [15]* [16]* [17] a method for ones. These architectures are often constructed with a performing 3-D object recognition directly from clutgreedy layer-by-layer method that models this idea. Deep tered scenes. Cresceptron is a cascade of many layers learning helps to disentangle these abstractions and pick similar to Neocognitron. But unlike Neocognitron which out which features are useful for learning.* [3] required the human programmer to hand-merge features,
For supervised learning tasks where label information is Cresceptron fully automatically learned an open numreadily available in training, deep learning promotes a ber of unsupervised features in each layer of the cascade principle which is very diﬀerent than traditional meth- where each feature is represented by a convolution kernel. ods of machine learning. That is, rather than focusing In addition, Cresceptron also segmented each learned obon feature engineering which is often labor-intensive and ject from a cluttered scene through back-analysis through varies from one task to another, deep learning methods is the network. Max-pooling, now often adopted by deep focused on end-to-end learning based on raw features. In neural networks (e.g., ImageNet tests), was ﬁrst used in other words, deep learning moves away from feature engi- Cresceptron to reduce the position resolution by a factor neering to a maximal extent possible. To accomplish end- of (2x2) to 1 through the cascade for better generalizato-end optimization starting with raw features and ending tion. Because of a great lack of understanding how the in labels, layered structures are often necessary. From brain autonomously wire its biological networks and the this perspective, we can regard the use of layered struc- computational cost by ANNs then, simpler models that tures to derive intermediate representations in deep learn- use task-speciﬁc handcrafted features such as Gabor ﬁling as a natural consequence of raw-feature-based end-to- ter and support vector machines (SVMs) were of popular end learning.* [1] Understanding the connection between choice of the ﬁeld in the 1990s and 2000s. the above two aspects of deep learning is important to ap- In the long history of speech recognition, both shalpreciate its use in several application areas, all involving low form and deep form (e.g., recurrent nets) of arsupervised learning tasks (e.g., supervised speech and im- tiﬁcial neural networks had been explored for many age recognition), as to be discussed in a later part of this years.* [18]* [19]* [20] But these methods never won over article. the non-uniform internal-handcrafting Gaussian mixture
Many deep learning algorithms are framed as unsuper- model/Hidden Markov model (GMM-HMM) technology vised learning problems. Because of this, these algo- based on generative models of speech trained discrim-

2.3. DEEP LEARNING IN ARTIFICIAL NEURAL NETWORKS inatively.* [21] A number of key diﬃculties had been methodologically analyzed, including gradient diminishing and weak temporal correlation structure in the neural predictive models.* [22]* [23] All these diﬃculties were in addition to the lack of big training data and big computing power in these early days. Most speech recognition researchers who understood such barriers hence subsequently moved away from neural nets to pursue generative modeling approaches until the recent resurgence of deep learning that has overcome all these diﬃculties. Hinton et al. and Deng et al. reviewed part of this recent history about how their collaboration with each other and then with cross-group colleagues ignited the renaissance of neural networks and initiated deep learning research and applications in speech recognition.* [24]* [25]* [26]* [27]
The term “deep learning”gained traction in the mid2000s after a publication by Geoﬀrey Hinton and Ruslan
Salakhutdinov showed how a many-layered feedforward neural network could be eﬀectively pre-trained one layer at a time, treating each layer in turn as an unsupervised restricted Boltzmann machine, then using supervised backpropagation for ﬁne-tuning.* [28] In 1992, Schmidhuber had already implemented a very similar idea for the more general case of unsupervised deep hierarchies of recurrent neural networks, and also experimentally shown its beneﬁts for speeding up supervised learning * [29]* [30]
Since the resurgence of deep learning, it has become part of many state-of-the-art systems in diﬀerent disciplines, particularly that of computer vision and automatic speech recognition (ASR). Results on commonly used evaluation sets such as TIMIT (ASR) and MNIST (image classiﬁcation) as well as a range of large vocabulary speech recognition tasks are constantly being improved with new applications of deep learning.* [24]* [31]* [32]
Currently, it has been shown that deep learning architectures in the form of convolutional neural networks have been nearly best performing;* [33]* [34] however, these are more widely used in computer vision than in ASR.
The real impact of deep learning in industry started in large-scale speech recognition around 2010. In late 2009,
Geoﬀ Hinton was invited by Li Deng to work with him and colleagues at Microsoft Research in Redmond to apply deep learning to speech recognition. They coorganized the 2009 NIPS Workshop on Deep Learning for Speech Recognition. The workshop was motivated by the limitations of deep generative models of speech, and the possibility that the big-compute, big-data era warranted a serious try of the deep neural net (DNN) approach. It was then (incorrectly) believed that pretraining of DNNs using generative models of deep belief net (DBN) would be the cure for the main diﬃculties of neural nets encountered during 1990's.* [26] However, soon after the research along this direction started at Microsoft Research, it was discovered that when large amounts of training data are used and especially when
DNNs are designed correspondingly with large, contextdependent output layers, dramatic error reduction oc-

15

curred over the then-state-of-the-art GMM-HMM and more advanced generative model-based speech recognition systems without the need for generative DBN pretraining, the ﬁnding veriﬁed subsequently by several other major speech recognition research groups * [24]* [35]
Further, the nature of recognition errors produced by the two types of systems was found to be characteristically diﬀerent,* [25]* [36] oﬀering technical insights into how to artfully integrate deep learning into the existing highly eﬃcient, run-time speech decoding system deployed by all major players in speech recognition industry. The history of this signiﬁcant development in deep learning has been described and analyzed in recent books.* [1]* [37]
Advances in hardware have also been an important enabling factor for the renewed interest of deep learning.
In particular, powerful graphics processing units (GPUs) are highly suited for the kind of number crunching, matrix/vector math involved in machine learning. GPUs have been shown to speed up training algorithms by orders of magnitude, bringing running times of weeks back to days.* [38]* [39]

2.3 Deep learning in artiﬁcial neural networks
Some of the most successful deep learning methods involve artiﬁcial neural networks. Artiﬁcial neural networks are inspired by the 1959 biological model proposed by Nobel laureates David H. Hubel & Torsten
Wiesel, who found two types of cells in the primary visual cortex: simple cells and complex cells. Many artiﬁcial neural networks can be viewed as cascading models
*
[15]* [16]* [17]* [40] of cell types inspired by these biological observations.
Fukushima's Neocognitron introduced convolutional neural networks partially trained by unsupervised learning while humans directed features in the neural plane. Yann LeCun et al. (1989) applied supervised backpropagation to such architectures.* [41] Weng et al.
(1992) published convolutional neural networks Cresceptron* [15]* [16]* [17] for 3-D object recognition from images of cluttered scenes and segmentation of such objects from images.
An obvious need for recognizing general 3-D objects is least shift invariance and tolerance to deformation.
Max-pooling appeared to be ﬁrst proposed by Cresceptron* [15]* [16] to enable the network to tolerate small-tolarge deformation in a hierarchical way while using convolution. Max-pooling helps, but still does not fully guarantee, shift-invariance at the pixel level.* [17]
With the advent of the back-propagation algorithm in the 1970s, many researchers tried to train supervised deep artiﬁcial neural networks from scratch, initially with little success. Sepp Hochreiter's diploma thesis of

16

CHAPTER 2. DEEP LEARNING

1991* [42]* [43] formally identiﬁed the reason for this failure in the “vanishing gradient problem,”which not only aﬀects many-layered feedforward networks, but also recurrent neural networks. The latter are trained by unfolding them into very deep feedforward networks, where a new layer is created for each time step of an input sequence processed by the network. As errors propagate from layer to layer, they shrink exponentially with the number of layers.

all other machine learning techniques on the old, famous
MNIST handwritten digits problem of Yann LeCun and colleagues at NYU.

Another method is the long short term memory (LSTM) network of 1997 by Hochreiter & Schmidhuber.* [44]
In 2009, deep multidimensional LSTM networks won three ICDAR 2009 competitions in connected handwriting recognition, without any prior knowledge about the three diﬀerent languages to be learned.* [45]* [46]

To break the barriers of weak AI represented by deep learning, it is necessary to go beyond the deep learning architectures because biological brains use both shallow and deep circuits as reported by brain anatomy* [58] in order to deal with the wide variety of invariance that the brain displays. Weng* [59] argued that the brain selfwires largely according to signal statistics and, therefore, a serial cascade cannot catch all major statistical dependencies. Fully guaranteed shift invariance for ANNs to deal with small and large natural objects in large cluttered scenes became true when the invariance went beyond shift, to extend to all ANN-learned concepts, such as location, type (object class label), scale, lighting, in the Developmental Networks (DNs)* [60] whose embodiments are Where-What Networks, WWN-1 (2008)* [61] through WWN-7 (2013).* [62]

As of 2011, the state of the art in deep learning feedforward networks alternates convolutional layers and maxpooling layers,* [53]* [54] topped by several pure classiﬁcation layers. Training is usually done without any unsupervised pre-training. Since 2011, GPU-based implementations* [53] of this approach won many pattern
To overcome this problem, several methods were pro- recognition contests, including the IJCNN 2011 Trafposed. One is Jürgen Schmidhuber's multi-level hi- ﬁc Sign Recognition Competition,* [55] the ISBI 2012 erarchy of networks (1992) pre-trained one level at a Segmentation of neuronal structures in EM stacks chaltime through unsupervised learning, ﬁne-tuned through lenge,* [56] and others. backpropagation.* [29] Here each level learns a com- Such supervised deep learning methods also were the pressed representation of the observations that is fed to ﬁrst artiﬁcial pattern recognizers to achieve humanthe next level. competitive performance on certain tasks.* [57]

Sven Behnke relied only on the sign of the gradient
(Rprop) when training his Neural Abstraction Pyramid* [47] to solve problems like image reconstruction and face localization.
Other methods also use unsupervised pre-training to structure a neural network, making it ﬁrst learn generally useful feature detectors. Then the network is trained further by supervised back-propagation to classify labeled data. The deep model of Hinton et al. (2006) involves learning the distribution of a high level representation using successive layers of binary or real-valued latent variables. It uses a restricted Boltzmann machine (Smolensky, 1986* [48]) to model each new layer of higher level features. Each new layer guarantees an increase on the lower-bound of the log likelihood of the data, thus improving the model, if trained properly. Once suﬃciently many layers have been learned the deep architecture may be used as a generative model by reproducing the data when sampling down the model (an “ancestral pass”) from the top level feature activations.* [49] Hinton reports that his models are eﬀective feature extractors over highdimensional, structured data.* [50]

2.4 Deep learning architectures
There are huge number of diﬀerent variants of deep architectures; however, most of them are branched from some original parent architectures. It is not always possible to compare the performance of multiple architectures all together, since they are not all implemented on the same data set. Deep learning is a fast-growing ﬁeld so new architectures, variants, or algorithms may appear every few weeks.

The Google Brain team led by Andrew Ng and Jeﬀ Dean created a neural network that learned to recognize higher2.4.1 Deep neural networks level concepts, such as cats, only from watching unlabeled images taken from YouTube videos.* [51] * [52]
A deep neural network (DNN) is an artiﬁcial neural
Other methods rely on the sheer processing power of network with multiple hidden layers of units between modern computers, in particular, GPUs. In 2010 it was the input and output layers.* [2]* [4] Similar to shallow shown by Dan Ciresan and colleagues* [38] in Jürgen ANNs, DNNs can model complex non-linear relationSchmidhuber's group at the Swiss AI Lab IDSIA that ships. DNN architectures, e.g., for object detection and despite the above-mentioned “vanishing gradient prob- parsing generate compositional models where the oblem,” superior processing power of GPUs makes plain ject is expressed as layered composition of image primthe back-propagation feasible for deep feedforward neural itives.* [63] The extra layers enable composition of feanetworks with many layers. The method outperformed tures from lower layers, giving the potential of modeling

2.4. DEEP LEARNING ARCHITECTURES

17

complex data with fewer units than a similarly performing to better local optima in comparison with other training shallow network.* [2] methods. However, these methods can be computationDNNs are typically designed as feedforward networks, ally expensive, especially when being used to train DNNs. but recent research has successfully applied the deep There are many training parameters to be considered with learning architecture to recurrent neural networks for ap- a DNN, such as the size (number of layers and number plications such as language modeling.* [64] Convolutional of units per layer), the learning rate and initial weights. deep neural networks (CNNs) are used in computer vi- Sweeping through the parameter space for optimal pasion where their success is well-documented.* [65] More rameters may not be feasible due to the cost in time and computational resources. Various 'tricks' such as using recently, CNNs have been applied to acoustic modeling for automatic speech recognition (ASR), where they have mini-batching (computing the gradient on several training examples at once rather than individual examples) shown success over previous models.* [34] For simplicity, *
[69] have been shown to speed up computation. The a look at training DNNs is given here. large processing throughput of GPUs has produced sigA DNN can be discriminatively trained with the standard niﬁcant speedups in training, due to the matrix and vector backpropagation algorithm. The weight updates can be computations required being well suited for GPUs.* [4] done via stochastic gradient descent using the following equation: 2.4.3 Deep belief networks
∆wij (t + 1) = ∆wij (t) + η

∂C
∂wij

Here, η is the learning rate, and C is the cost function. The choice of the cost function depends on factors such as the learning type (supervised, unsupervised, reinforcement, etc.) and the activation function.
For example, when performing supervised learning on a multiclass classiﬁcation problem, common choices for the activation function and cost function are the softmax function and cross entropy function, respectively. The exp(xj ) softmax function is deﬁned as pj = ∑ exp(xk ) where pj k represents the class probability and xj and xk represent the total input to units j and k respectively. Cross en∑ tropy is deﬁned as C = − j dj log(pj ) where dj represents the target probability for output unit j and pj is the probability output for j after applying the activation function.* [66]

Main article: Deep belief network
A deep belief network (DBN) is a probabilistic,

Hidden units
Visible units

h3

v3 h4 v1 h1 v2 h2 2.4.2

Issues with deep neural networks

As with ANNs, many issues can arise with DNNs if they are naively trained. Two common issues are overﬁtting A restricted Boltzmann machine (RBM) with fully connected visible and hidden units. Note there are no hidden-hidden or visibleand computation time. visible connections.

DNNs are prone to overﬁtting because of the added layers of abstraction, which allow them to model rare dependencies in the training data. Regularization methods such as weight decay ( ℓ2 -regularization) or sparsity
( ℓ1 -regularization) can be applied during training to help combat overﬁtting.* [67] A more recent regularization method applied to DNNs is dropout regularization.
In dropout, some number of units are randomly omitted from the hidden layers during training. This helps to break the rare dependencies that can occur in the training data * [68]

generative model made up of multiple layers of hidden units. It can be looked at as a composition of simple learning modules that make up each layer.* [70]

A DBN can be used for generatively pre-training a DNN by using the learned weights as the initial weights. Backpropagation or other discriminative algorithms can then be applied for ﬁne-tuning of these weights. This is particularly helpful in situations where limited training data is available, as poorly initialized weights can have signiﬁcant impact on the performance of the ﬁnal model. These
Backpropagation and gradient descent have been the pre- pre-trained weights are in a region of the weight space that ferred method for training these structures due to the is closer to the optimal weights (as compared to just ranease of implementation and their tendency to converge dom initialization). This allows for both improved mod-

18

CHAPTER 2. DEEP LEARNING

eling capability and faster convergence of the ﬁne-tuning a training vector and values for the units in the alreadyphase.* [71] trained RBM layers are assigned using the current weights
A DBN can be eﬃciently trained in an unsupervised, and biases. The ﬁnal layer of the already-trained layers is layer-by-layer manner where the layers are typically made used as input to the new RBM. The new RBM is then of restricted Boltzmann machines (RBM). A description trained with the procedure above, and then this whole of training a DBN via RBMs is provided below. An RBM process can be repeated until some desired stopping cri* is an undirected, generative energy-based model with an terion is met. [2] input layer and single hidden layer. Connections only exist between the visible units of the input layer and the hidden units of the hidden layer; there are no visible-visible or hidden-hidden connections.

Despite the approximation of CD to maximum likelihood being very crude (CD has been shown to not follow the gradient of any function), empirical results have shown it to be an eﬀective method for use with training deep
*
The training method for RBMs was initially proposed architectures. [69] by Geoﬀrey Hinton for use with training “Product of Expert”models and is known as contrastive diver2.4.4 Convolutional neural networks gence (CD).* [72] CD provides an approximation to the maximum likelihood method that would ideally be apMain article: Convolutional neural network plied for learning the weights of the RBM.* [69]* [73]
In training a single RBM, weight updates are performed with gradient ascent via the following equation: ∆wij (t+
1) = wij (t) + η ∂ log(p(v)) . Here, p(v) is the prob∂wij ability of a visible vector, which is given by p(v) =
∑ −E(v,h)
1
. Z is the partition function (used for he Z normalizing) and E(v, h) is the energy function assigned to the state of the network. A lower energy indicates the network is in a more
“desirable”
conﬁguration. The gradi∂ log(p(v)) ent ∂wij has the simple form ⟨vi hj ⟩data −⟨vi hj ⟩model where ⟨· · · ⟩p represent averages with respect to distribution p . The issue arises in sampling ⟨vi hj ⟩model as this requires running alternating Gibbs sampling for a long time. CD replaces this step by running alternating Gibbs sampling for n steps (values of n = 1 have empirically been shown to perform well). After n steps, the data is sampled and that sample is used in place of ⟨vi hj ⟩model .
The CD procedure works as follows:* [69]
1. Initialize the visible units to a training vector.
2. Update the hidden units in parallel given the visible
∑
units: p(hj = 1 | V) = σ(bj + i vi wij ) . σ represents the sigmoid function and bj is the bias of hj .
3. Update the visible units in parallel given the hidden
∑
units: p(vi = 1 | H) = σ(ai + j hj wij ) . ai is the bias of vi . This is called the “reconstruction” step. A CNN is composed of one or more convolutional layers with fully connected layers (matching those in typical artiﬁcial neural networks) on top. It also uses tied weights and pooling layers. This architecture allows CNNs to take advantage of the 2D structure of input data. In comparison with other deep architectures, convolutional neural networks are starting to show superior results in both image and speech applications. They can also be trained with standard backpropagation. CNNs are easier to train than other regular, deep, feed-forward neural networks and have many fewer parameters to estimate, making them a highly attractive architecture to use.* [74]

2.4.5 Convolutional Deep Belief Networks
A recent achievement in deep learning is from the use of convolutional deep belief networks (CDBN). A CDBN is very similar to normal Convolutional neural network in terms of its structure. Therefore, like CNNs they are also able to exploit the 2D structure of images combined with the advantage gained by pre-training in Deep belief network. They provide a generic structure which can be used in many image and signal processing tasks and can be trained in a way similar to that for Deep Belief Networks. Recently, many benchmark results on standard image datasets like CIFAR * [75] have been obtained using CDBNs.* [76]

4. Reupdate the hidden units in parallel given the re- 2.4.6 Deep Boltzmann Machines constructed visible units using the same equation as
A Deep Boltzmann Machine (DBM) is a type of biin step 2. nary pairwise Markov random ﬁeld (undirected proba5. Perform the weight update: ∆wij ∝ ⟨vi hj ⟩data − bilistic graphical models) with multiple layers of hidden
⟨vi hj ⟩reconstruction . random variables. It is a network of symmetrically coupled stochastic binary units. It comprises a set of visible
Once an RBM is trained, another RBM can be“stacked” units ν ∈ {0, 1}D , and a series of layers of hidden units atop of it to create a multilayer model. Each time another h(1) ∈ {0, 1}F1 , h(2) ∈ {0, 1}F2 , . . . , h(L) ∈ {0, 1}FL
RBM is stacked, the input visible layer is initialized to . There is no connection between the units of the same

2.4. DEEP LEARNING ARCHITECTURES

19

layer (like RBM). For the DBM, we can write the proba- y, where θ = {W , b} , W is the weight matrix and b is bility which is assigned to vector ν as: an oﬀset vector (bias). On the contrary a decoder maps back the
∑ ∑ W (1) νi h(1) +∑ W (2) h(1) h(2) +∑ W (3) h(2) h(3) hidden representation y to the reconstructed in1 m , jl lm j j jl l lm l p(ν) = Z h e ij ij put z via gθ . The whole process of auto encoding is to where h = {h(1) , h(2) , h(3) } are the set of hidden units, compare this reconstructed input to the original and try and θ = {W (1) , W (2) , W (3) } are the model parame- to minimize this error to make the reconstructed value as ters, representing visible-hidden and hidden-hidden sym- close as possible to the original. metric interaction, since they are undirected links. As it is clear by setting W (2) = 0 and W (3) = 0 the network becomes the well-known Restricted Boltzmann machine.* [77]

In stacked denoising auto encoders, the partially corrupted output is cleaned (denoised). This fact has been introduced in * [81] with a speciﬁc approach to good representation, a good representation is one that can be obtained
There are several reasons which motivate us to take ad- robustly from a corrupted input and that will be useful for vantage of deep Boltzmann machine architectures. Like recovering the corresponding clean input. Implicit in this
DBNs, they beneﬁt from the ability of learning complex deﬁnition are the ideas of and abstract internal representations of the input in tasks
• The higher level representations are relatively stable such as object or speech recognition, with the use of limand robust to the corruption of the input; ited number of labeled data to ﬁne-tune the representations built based on a large supply of unlabeled sensory in• It is required to extract features that are useful for put data. However, unlike DBNs and deep convolutional representation of the input distribution. neural networks, they adopt the inference and training procedure in both directions, bottom-up and top-down pass, which enable the DBMs to better unveil the rep- The algorithm consists of multiple steps; starts by a
˜
x resentations of the ambiguous and complex input struc- stochastic mapping of x to x through qD (˜ |x) , this is
*
*
˜
the corrupting step. Then the corrupted input x passes tures, [78] . [79] through a basic auto encoder process and is mapped to a
Since the exact maximum likelihood learning is intractable hidden representation y = f (˜ ) = s(W x + b) . From
˜
θ x for the DBMs, we may perform the approximate max- this hidden representation we can reconstruct z = g (y) θ imum likelihood learning. There is another possibility, . In the last stage a minimization algorithm is done in orto use mean-ﬁeld inference to estimate data-dependent der to have a z as close as possible to uncorrupted input x expectations, incorporation with a Markov chain Monte . The reconstruction error L (x, z) might be either the
H
Carlo (MCMC) based stochastic approximation technique cross-entropy loss with an aﬃne-sigmoid decoder, or the to approximate the expected suﬃcient statistics of the squared error loss with an aﬃne decoder.* [81] model.* [77]
In order to make a deep architecture, auto encoders stack
We can see the diﬀerence between DBNs and DBM. In one on top of another. Once the encoding function f θ DBNs, the top two layers form a restricted Boltzmann of the ﬁrst denoising auto encoder is learned and used to machine which is an undirected graphical model, but the uncorrupt the input (corrupted input), we can train the lower layers form a directed generative model. second level.* [81]
Apart from all the advantages of DBMs discussed so far, Once the stacked auto encoder is trained, its output might they have a crucial disadvantage which limits the per- be used as the input to a supervised learning algorithm formance and functionality of this kind of architecture. such as support vector machine classiﬁer or a multiclass
The approximate inference, which is based on mean- logistic regression.* [81] ﬁeld method, is about 25 to 50 times slower than a single bottom-up pass in DBNs. This time consuming task make the joint optimization, quite impractical for large 2.4.8 Deep Stacking Networks data sets, and seriously restricts the use of DBMs in tasks such as feature representations (the mean-ﬁeld inference One of the deep architectures recently introduced in* [82] have to be performed for each new test input).* [80] which is based on building hierarchies with blocks of simpliﬁed neural network modules, is called deep convex network. They are called “convex”because of the
2.4.7 Stacked (Denoising) Auto-Encoders formulation of the weights learning problem, which is a convex optimization problem with a closed-form soluThe auto encoder idea is motivated by the concept of good tion. The network is also called the deep stacking network representation. For instance for the case of classiﬁer it is (DSN),* [83] emphasizing on this fact that a similar mechpossible to deﬁne that a good representation is one that will anism as the stacked generalization is used.* [84] yield a better performing classiﬁer.
The DSN blocks, each consisting of a simple, easy-toAn encoder is referred to a deterministic mapping fθ that learn module, are stacked to form the overall deep nettransforms an input vector x into hidden representation work. It can be trained block-wise in a supervised fash-

20

CHAPTER 2. DEEP LEARNING

ion without the need for back-propagation for the entire scale up the design to larger (deeper) architectures and blocks.* [85] data sets.
As designed in * [82] each block consists of a simpliﬁed
MLP with a single hidden layer. It comprises a weight matrix U as the connection between the logistic sigmoidal units of the hidden layer h to the linear output layer y, and a weight matrix W which connects each input of the blocks to their respective hidden layers. If we assume that the target vectors t be arranged to form the columns of T (the target matrix), let the input data vectors x be arranged to form the columns of X, let H = σ(W T X) denote the matrix of hidden units, and assume the lowerlayer weights W are known (training layer-by-layer). The function performs the element-wise logistic sigmoid operation. Then learning the upper-layer weight matrix U given other weights in the network can be formulated as a convex optimization problem:

The basic architecture is suitable for diverse tasks such as classiﬁcation and regression.

In each block an estimate of the same ﬁnal label class y is produced, then this estimated label concatenated with original input to form the expanded input for the upper block. In contrast with other deep architectures, such as
DBNs, the goal is not to discover the transformed feature representation. Regarding the structure of the hierarchy of this kind of architecture, it makes the parallel training straightforward as the problem is naturally a batch-mode optimization one. In purely discriminative tasks DSN performance is better than the conventional DBN.* [83]

capacity to the architecture using additional terms in the energy function. One of these terms enable model to form a conditional distribution of the spike variables by means of marginalizing out the slab variables given an observation.

2.4.10 Spike-and-Slab RBMs (ssRBMs)
The need for real-valued inputs which are employed in
Gaussian RBMs (GRBMs), motivates scientists seeking new methods. One of these methods is the spike and slab
RBM (ssRBMs), which models continuous-valued inputs with strictly binary latent variables.* [90]

Similar to basic RBMs and its variants, the spike and slab RBM is a bipartite graph. Like GRBM, the visible units (input) are real-valued. The diﬀerence arises in the hidden layer, where each hidden unit come along with a binary spike variable and real-valued slab variable.
These terms (spike and slab) come from the statistics litmin f = ||U T H − T ||2 , erature,* [91] and refer to a prior including a mixture of
F
UT two components. One is a discrete probability mass at which has a closed-form solution. The input to the ﬁrst zero called spike, and the other is a density over continu*
*
block X only contains the original data, however in the ous domain. [92] [92] upper blocks in addition to this original (raw) data there There is also an extension of the ssRBM model, which is a copy of the lower-block(s) output y. is called µ-ssRBM. This variant provides extra modeling

2.4.9

Tensor Deep Stacking Networks (TDSN)

This architecture is an extension of the DSN. It improves the DSN in two important ways, using the higher order information by means of covariance statistics and transforming the non-convex problem of the lower-layer to a convex sub-problem of the upper-layer.* [86]
Unlike the DSN, the covariance statistics of the data is employed using a bilinear mapping from two distinct sets of hidden units in the same layer to predictions via a thirdorder tensor.
The scalability and parallelization are the two important factors in the learning algorithms which are not considered seriously in the conventional DNNs.* [87]* [88]* [89]
All the learning process for the DSN (and TDSN as well) is done on a batch-mode basis so as to make the parallelization possible on a cluster of CPU or GPU nodes.* [82]* [83] Parallelization gives the opportunity to

2.4.11 Compound
Models

Hierarchical-Deep

The class architectures called compound HD models, where HD stands for Hierarchical-Deep are structured as a composition of non-parametric Bayesian models with deep networks. The features, learned by deep architectures such as DBNs,* [93] DBMs,* [78] deep auto encoders,* [94] convolutional variants,* [95]* [96] ssRBMs,* [92] deep coding network,* [97] DBNs with sparse feature learning,* [98] recursive neural networks,* [99] conditional DBNs,* [100] denoising auto encoders,* [101] are able to provide better representation for more rapid and accurate classiﬁcation tasks with high-dimensional training data sets. However, they are not quite powerful in learning novel classes with few examples, themselves. In these architectures, all units through the network are involved in the representation of the input (distributed representations), and they have to be adjusted together (high degree of freedom). However, if we limit the degree of freedom, we make it easier for the model to learn new classes out of few training samples (less parameters to learn). Hierarchical Bayesian
(HB) models, provide learning from few examples, for example * [102]* [103]* [104]* [105]* [106] for computer

2.5. APPLICATIONS

21

vision, statistics, and cognitive science.

It is also possible to extend the DPCN to form a
*
Compound HD architectures try to integrate both charac- convolutional network. [108] teristics of HB and deep networks. The compound HDPDBM architecture, a hierarchical Dirichlet process (HDP) 2.4.13 Deep Kernel Machines as a hierarchical model, incorporated with DBM architecture. It is a full generative model, generalized from The Multilayer Kernel Machine (MKM) as introduced in abstract concepts ﬂowing through the layers of the model, * [109] is a way of learning highly nonlinear functions which is able to synthesize new examples in novel classes with the iterative applications of weakly nonlinear kerthat look reasonably natural. Note that all the levels nels. They use the kernel principal component analyare learned jointly by maximizing a joint log-probability sis (KPCA), in,* [110] as method for unsupervised greedy score.* [107] layer-wise pre-training step of the deep learning architecConsider a DBM with three hidden layers, the probability ture. of a visible input ν is:
Layer l + 1 -th learns the representation of the previous
∑ ∑ W (1) νi h1 +∑ W (2) h1 h2 +∑ W (3) h2 h3 l , extracting the n principal component (PC) of the
1
layer j j l jl lm l jl lm l m , p(ν, ψ) = Z h e ij ij projection layer l output in the feature domain induced where h = {h(1) , h(2) , h(3) } are the set of hidden units, by the kernel. For the sake of dimensionality reduction and ψ = {W (1) , W (2) , W (3) } are the model parame- of the updated representation in each layer, a supervised ters, representing visible-hidden and hidden-hidden sym- strategy is proposed to select the best informative features metric interaction terms. among the ones extracted by KPCA. The process is:
After a DBM model has been learned, we have an undirected model that deﬁnes the joint distribution
P (ν, h1 , h2 , h3 ) . One way to express what has been learned is the conditional model P (ν, h1 , h2 |h3 ) and a prior term P (h3 ) .

• ranking the nl features according to their mutual information with the class labels;
• for diﬀerent values of K and ml ∈ {1, . . . , nl } , compute the classiﬁcation error rate of a K-nearest neighbor (K-NN) classiﬁer using only the ml most informative features on a validation set;

The part P (ν, h1 , h2 |h3 ) , represents a conditional DBM model, which can be viewed as a two-layer DBM but with bias terms given by the states of h3 :
P (ν, h1 , h2 |h3 ) =

2.4.12

1
Z(ψ,h3 ) e

∑ ij (1)

Wij νi h1 + j ∑ jl (2)

Wjl h1 h2 + j l

Deep Coding Networks

There are several advantages to having a model which can actively update itself to the context in data. One of these methods arises from the idea to have a model which is able to adjust its prior knowledge dynamically according to the context of the data. Deep coding network (DPCN) is a predictive coding scheme where top-down information is used to empirically adjust the priors needed for the bottom-up inference procedure by means of a deep locally-connected generative model. This is based on extracting sparse features out of time-varying observations using a linear dynamical model. Then, a pooling strategy is employed in order to learn invariant feature representations. Similar to other deep architectures, these blocks are the building elements of a deeper architecture where greedy layer-wise unsupervised learning are used. Note that the layers constitute a kind of Markov chain such that the states at any layer are only dependent on the succeeding and preceding layers.

∑

• the value of ml with which the classiﬁer has reached
(3)
Wlm h2 h3 l m. the lowest error rate determines the number of features to retain.

lm

There are some drawbacks in using the KPCA method as the building cells of an MKM.
Another, more straightforward method of integrating kernel machine into the deep learning architecture was developed by Microsoft researchers for spoken language understanding applications.* [111] The main idea is to use a kernel machine to approximate a shallow neural net with an inﬁnite number of hidden units, and then to use the stacking technique to splice the output of the kernel machine and the raw input in building the next, higher level of the kernel machine. The number of the levels in this kernel version of the deep convex network is a hyperparameter of the overall system determined by cross validation.

2.4.14 Deep Q-Networks

This is the latest class of deep learning models targeted for reinforcement learning, published in February 2015
*
*
Deep predictive coding network (DPCN) [108] predicts in Nature [112] the representation of the layer, by means of a top-down approach using the information in upper layer and also temporal dependencies from the previous states, it is 2.5 Applications called 22

2.5.1

CHAPTER 2. DEEP LEARNING

Automatic speech recognition

The results shown in the table below are for automatic speech recognition on the popular TIMIT data set. This is a common data set used for initial evaluations of deep learning architectures. The entire set contains 630 speakers from eight major dialects of American English, with each speaker reading 10 diﬀerent sentences.* [113] Its small size allows many diﬀerent conﬁgurations to be tried eﬀectively with it. More importantly, the TIMIT task concerns phone-sequence recognition, which, unlike word-sequence recognition, permits very weak
“language
models” thus the weaknesses in acoustic modeling asand pects of speech recognition can be more easily analyzed.
It was such analysis on TIMIT contrasting the GMM (and other generative models of speech) vs. DNN models carried out by Li Deng and collaborators around 20092010 that stimulated early industrial investment on deep learning technology for speech recognition from small to large scales,* [25]* [36] eventually leading to pervasive and dominant uses of deep learning in speech recognition industry. That analysis was carried out with comparable performance (less than 1.5% in error rate) between discriminative DNNs and generative models. The error rates presented below, including these early results and measured as percent phone error rates (PER), have been summarized over a time span of the past 20 years:
Extension of the success of deep learning from TIMIT to large vocabulary speech recognition occurred in 2010 by industrial researchers, where large output layers of the DNN based on context dependent HMM states constructed by decision trees were adopted.* [116]* [117] See comprehensive reviews of this development and of the state of the art as of October 2014 in the recent Springer book from Microsoft Research.* [37] See also the related background of automatic speech recognition and the impact of various machine learning paradigms including notably deep learning in a recent overview article.* [118]
One fundamental principle of deep learning is to do away with hand-crafted feature engineering and to use raw features. This principle was ﬁrst explored successfully in the architecture of deep autoencoder on the “raw”spectrogram or linear ﬁlter-bank features,* [119] showing its superiority over the Mel-Cepstral features which contain a few stages of ﬁxed transformation from spectrograms.
The true “raw”features of speech, waveforms, have more recently been shown to produce excellent largerscale speech recognition results.* [120]

els; 5) Multi-task and transfer learning by DNNs and related deep models; 6) Convolution neural networks and how to design them to best exploit domain knowledge of speech; 7) Recurrent neural network and its rich LSTM variants; 8) Other types of deep models including tensor-based models and integrated deep generative/discriminative models.
Large-scale automatic speech recognition is the ﬁrst and the most convincing successful case of deep learning in the recent history, embraced by both industry and academic across the board. Between 2010 and 2014, the two major conferences on signal processing and speech recognition, IEEE-ICASSP and Interspeech, have seen near exponential growth in the numbers of accepted papers in their respective annual conference papers on the topic of deep learning for speech recognition. More importantly, all major commercial speech recognition systems (e.g., Microsoft Cortana, Xbox, Skype Translator,
Google Now, Apple Siri, Baidu and iFlyTek voice search, and a range of Nuance speech products, etc.) nowadays are based on deep learning methods.* [1]* [121]* [122]
See also the recent media interview with the CTO of Nuance Communications.* [123]
The wide-spreading success in speech recognition achieved by 2011 was followed shortly by large-scale image recognition described next.

2.5.2 Image recognition
A common evaluation set for image classiﬁcation is the
MNIST database data set. MNIST is composed of handwritten digits and includes 60000 training examples and
10000 test examples. Similar to TIMIT, its small size allows multiple conﬁgurations to be tested. A comprehensive list of results on this set can be found in.* [124] The current best result on MNIST is an error rate of 0.23%, achieved by Ciresan et al. in 2012.* [125]

The real impact of deep learning in image or object recognition, one major branch of computer vision, was felt in the fall of 2012 after the team of Geoﬀ Hinton and his students won the large-scale ImageNet competition by a signiﬁcant margin over the then-state-of-the-art shallow machine learning methods. The technology is based on
20-year-old deep convolutional nets, but with much larger scale on a much larger task, since it had been learned that deep learning works quite well on large-scale speech recognition. In 2013 and 2014, the error rate on the ImSince the initial successful debut of DNNs for speech ageNet task using deep learning was further reduced at a recognition around 2009-2011, there has been huge rapid pace, following a similar trend in large-scale speech progress made. This progress (as well as future direc- recognition. tions) has been summarized into the following eight ma- As in the ambitious moves from automatic speech recogjor areas:* [1]* [27]* [37] 1) Scaling up/out and speedup nition toward automatic speech translation and underDNN training and decoding; 2) Sequence discriminative standing, image classiﬁcation has recently been extended training of DNNs; 3) Feature processing by deep mod- to the more ambitious and challenging task of automatic els with solid understanding of the underlying mecha- image captioning, in which deep learning is the essential nisms; 4) Adaptation of DNNs and of related deep mod- underlying technology. * [126] * [127] * [128] * [129]

2.6. DEEP LEARNING IN THE HUMAN BRAIN

23

One example application is a car computer said to be illustrating suitability of the method for CRM automatrained with deep learning, which may be able to let cars tion. A neural network was used to approximate the value interpret 360° camera views.* [130] of possible direct marketing actions over the customer state space, deﬁned in terms of RFM variables. The estimated value function was shown to have a natural inter2.5.3 Natural language processing pretation as CLV (customer lifetime value).* [151]
Neural networks have been used for implementing language models since the early 2000s.* [131] Key techniques in this ﬁeld are negative sampling* [132] and word embedding. A word embedding, such as word2vec, can be thought of as a representational layer in a deep learning architecture transforming an atomic word into a positional representation of the word relative to other words in the dataset; the position is represented as a point in a vector space. Using a word embedding as an input layer to a recursive neural network (RNN) allows for the training of the network to parse sentences and phrases using an eﬀective compositional vector grammar. A compositional vector grammar can be thought of as probabilistic context free grammar (PCFG) implemented by a recursive neural network.* [133] Recursive autoencoders built atop word embeddings have been trained to assess sentence similarity and detect paraphrasing.* [133] Deep neural architectures have achieved state-of-the-art results in many tasks in natural language processing, such as constituency parsing,* [134] sentiment analysis,* [135] information retrieval,* [136] * [137] machine translation, * [138] * [139] contextual entity linking, * [140] and other areas of NLP.
*
[141]

2.5.4

Drug discovery and toxicology

The pharmaceutical industry faces the problem that a large percentage of candidate drugs fail to reach the market. These failures of chemical compounds are caused by insuﬃcient eﬃcacy on the biomolecular target (on-target eﬀect), undetected and undesired interactions with other biomolecules (oﬀ-target eﬀects), or unanticipated toxic eﬀects.* [142]* [143] In 2012 a team led by George Dahl won the“Merck Molecular Activity
Challenge”
using multi-task deep neural networks to predict the biomolecular target of a compound.* [144]* [145]
In 2014 Sepp Hochreiter's group used Deep Learning to detect oﬀ-target and toxic eﬀects of environmental chemicals in nutrients, household products and drugs and won the “Tox21 Data Challenge”of NIH, FDA and
NCATS.* [146]* [147] These impressive successes show
Deep Learning may be superior to other virtual screening methods.* [148]* [149] Researchers from Google and
Stanford enhanced Deep Learning for drug discovery by combining data from a variety of sources.* [150]

2.6 Deep learning in the human brain Computational deep learning is closely related to a class of theories of brain development (speciﬁcally, neocortical development) proposed by cognitive neuroscientists in the early 1990s.* [152] An approachable summary of this work is Elman, et al.'s 1996 book “Rethinking Innateness”* [153] (see also: Shrager and Johnson;* [154] Quartz and Sejnowski * [155]). As these developmental theories were also instantiated in computational models, they are technical predecessors of purely computationally-motivated deep learning models. These developmental models share the interesting property that various proposed learning dynamics in the brain (e.g., a wave of nerve growth factor) conspire to support the self-organization of just the sort of inter-related neural networks utilized in the later, purely computational deep learning models; and such computational neural networks seem analogous to a view of the brain's neocortex as a hierarchy of ﬁlters in which each layer captures some of the information in the operating environment, and then passes the remainder, as well as modiﬁed base signal, to other layers further up the hierarchy. This process yields a self-organizing stack of transducers, well-tuned to their operating environment. As described in The New York
Times in 1995: "...the infant's brain seems to organize itself under the inﬂuence of waves of so-called trophicfactors ... diﬀerent regions of the brain become connected sequentially, with one layer of tissue maturing before another and so on until the whole brain is mature.”
*
[156]

The importance of deep learning with respect to the evolution and development of human cognition did not escape the attention of these researchers. One aspect of human development that distinguishes us from our nearest primate neighbors may be changes in the timing of development.* [157] Among primates, the human brain remains relatively plastic until late in the post-natal period, whereas the brains of our closest relatives are more completely formed by birth. Thus, humans have greater access to the complex experiences aﬀorded by being out in the world during the most formative period of brain development. This may enable us to “tune in”to rapidly changing features of the environment that other
2.5.5 Customer relationship management animals, more constrained by evolutionary structuring of their brains, are unable to take account of. To the exRecently success has been reported with application of tent that these changes are reﬂected in similar timing deep reinforcement learning in direct marketing settings, changes in hypothesized wave of cortical development,

24 they may also lead to changes in the extraction of information from the stimulus environment during the early self-organization of the brain. Of course, along with this ﬂexibility comes an extended period of immaturity, during which we are dependent upon our caretakers and our community for both support and training. The theory of deep learning therefore sees the coevolution of culture and cognition as a fundamental condition of human evolution.* [158]

2.7 Commercial activity
Deep learning is often presented as a step towards realising strong AI* [159] and thus many organizations have become interested in its use for particular applications.
Most recently, in December 2013, Facebook announced that it hired Yann LeCun to head its new artiﬁcial intelligence (AI) lab that will have operations in California,
London, and New York. The AI lab will be used for developing deep learning techniques that will help Facebook do tasks such as automatically tagging uploaded pictures with the names of the people in them.* [160]
In March 2013, Geoﬀrey Hinton and two of his graduate students, Alex Krizhevsky and Ilya Sutskever, were hired by Google. Their work will be focused on both improving existing machine learning products at Google and also help deal with the growing amount of data Google has.
Google also purchased Hinton's company, DNNresearch.
In 2014 Google also acquired DeepMind Technologies, a
British start-up that developed a system capable of learning how to play Atari video games using only raw pixels as data input.
Baidu hired Andrew Ng to head their new Silicon Valley based research lab focusing on deep learning.

2.8 Criticism and comment
Given the far-reaching implications of artiﬁcial intelligence coupled with the realization that deep learning is emerging as one of its most powerful techniques, the subject is understandably attracting both criticism and comment, and in some cases from outside the ﬁeld of computer science itself.
A main criticism of deep learning concerns the lack of theory surrounding many of the methods. Most of the learning in deep architectures is just some form of gradient descent. While gradient descent has been understood for a while now, the theory surrounding other algorithms, such as contrastive divergence is less clear (i.e.,
Does it converge? If so, how fast? What is it approximating?). Deep learning methods are often looked at as a black box, with most conﬁrmations done empirically, rather than theoretically.

CHAPTER 2. DEEP LEARNING
Others point out that deep learning should be looked at as a step towards realizing strong AI, not as an allencompassing solution. Despite the power of deep learning methods, they still lack much of the functionality needed for realizing this goal entirely. Research psychologist Gary Marcus has noted that:
“Realistically, deep learning is only part of the larger challenge of building intelligent machines. Such techniques lack ways of representing causal relationships (...) have no obvious ways of performing logical inferences, and they are also still a long way from integrating abstract knowledge, such as information about what objects are, what they are for, and how they are typically used. The most powerful A.I. systems, like Watson (...) use techniques like deep learning as just one element in a very complicated ensemble of techniques, ranging from the statistical technique of Bayesian inference to deductive reasoning.”* [161]
To the extent that such a viewpoint implies, without intending to, that deep learning will ultimately constitute nothing more than the primitive discriminatory levels of a comprehensive future machine intelligence, a recent pair of speculations regarding art and artiﬁcial intelligence* [162] oﬀers an alternative and more expansive outlook. The ﬁrst such speculation is that it might be possible to train a machine vision stack to perform the sophisticated task of discriminating between “old master”and amateur ﬁgure drawings; and the second is that such a sensitivity might in fact represent the rudiments of a non-trivial machine empathy. It is suggested, moreover, that such an eventuality would be in line with both anthropology, which identiﬁes a concern with aesthetics as a key element of behavioral modernity, and also with a current school of thought which suspects that the allied phenomenon of consciousness, formerly thought of as a purely high-order phenomenon, may in fact have roots deep within the structure of the universe itself.
Some currently popular and successful deep learning architectures display certain problematical behaviors* [163]
(e.g. conﬁdently classifying random data as belonging to a familiar category of nonrandom images;* [164] and misclassifying miniscule perturbations of correctly classiﬁed images * [165]). The creator of OpenCog, Ben Goertzel hypothesized * [163] that these behaviors are tied with limitations in the internal representations learned by these architectures, and that these same limitations would inhibit integration of these architectures into heterogeneous multi-component AGI architectures. It is suggested that these issues can be worked around by developing deep learning architectures that internally form states homologous to image-grammar * [166] decompositions of observed entities and events.* [163] Learning a grammar
(visual or linguistic) from training data would be equivalent to restricting the system to commonsense reasoning which operates on concepts in terms of production rules of the grammar, and is a basic goal of both human language acquisition * [167] and A.I. (Also see Grammar in-

2.11. REFERENCES duction * [168])

2.9 Deep learning software libraries
• Torch
• Theano
• Deeplearning4j, distributed deep learning for the
JVM. Parallel GPUs.
• ND4J
• NVIDIA cuDNN library of accelerated primitives for deep neural networks.
• DeepLearnToolbox, Matlab/Octave toolbox for deep learning

25

2.11 References
[1] L. Deng and D. Yu (2014) “Deep Learning: Methods and Applications”http://research.microsoft.com/pubs/
209355/DeepLearning-NowPublishing-Vol7-SIG-039.
pdf
[2] Bengio, Yoshua (2009). “Learning Deep Architectures for AI” (PDF). Foundations and Trends in Machine
Learning 2 (1).
[3] Y. Bengio, A. Courville, and P. Vincent., Representation
“
Learning: A Review and New Perspectives,”
IEEE Trans.
PAMI, special issue Learning Deep Architectures, 2013
[4] J. Schmidhuber, Deep Learning in Neural Networks: An
“
Overview”http://arxiv.org/abs/1404.7828, 2014
[5] Patrick Glauner (2015), Comparison of Training Methods for Deep Neural Networks, arXiv:1504.06825
[6] Song, Hyun Ah, and Soo-Young Lee.“Hierarchical Representation Using NMF.”
Neural Information Processing.
Springer Berlin Heidelberg, 2013.

• convnetjs, deep learning library in Javascript. Contains online demos.

[7] Olshausen, Bruno A. “Emergence of simple-cell receptive ﬁeld properties by learning a sparse code for natural images.”Nature 381.6583 (1996): 607-609.

• Gensim a toolkit for natural language processing; includes word2vec

[8] Ronan Collobert (May 6, 2011). “Deep Learning for
Eﬃcient Discriminative Parsing”. videolectures.net. Ca.
7:45.

• Caﬀe

2.10 See also
• Unsupervised learning
• Graphical model
• Feature learning
• Sparse coding
• Compressed Sensing
• Connectionism
• Self-organizing map
• Principal component analysis
• Applications of artiﬁcial intelligence
• List of artiﬁcial intelligence projects
• Extreme Learning Machines
• Reservoir computing
• Liquid state machine
• Echo state network

[9] Gomes, Lee (20 October 2014). “Machine-Learning
Maestro Michael Jordan on the Delusions of Big Data and
Other Huge Engineering Eﬀorts”. IEEE Spectrum.
[10] Fukushima, K. (1980).“Neocognitron: A self-organizing neural network model for a mechanism of pattern recognition unaﬀected by shift in position”. Biol. Cybern 36:
193–202. doi:10.1007/bf00344251.
[11] P. Werbos.,“Beyond Regression: New Tools for Prediction and Analysis in the Behavioral Sciences,”
PhD thesis,
Harvard University, 1974.
[12] LeCun et al., “Backpropagation Applied to Handwritten
Zip Code Recognition,”
Neural Computation, 1, pp. 541–
551, 1989.
[13] S. Hochreiter., "Untersuchungen zu dynamischen neuronalen Netzen,”Diploma thesis. Institut f. Informatik,
Technische Univ. Munich. Advisor: J. Schmidhuber,
1991.
[14] S. Hochreiter et al., “Gradient ﬂow in recurrent nets: the diﬃculty of learning long-term dependencies,”In S. C.
Kremer and J. F. Kolen, editors, A Field Guide to Dynamical Recurrent Neural Networks. IEEE Press, 2001.
[15] J. Weng, N. Ahuja and T. S. Huang, "Cresceptron: a selforganizing neural network which grows adaptively,”
Proc.
International Joint Conference on Neural Networks, Baltimore, Maryland, vol I, pp. 576-581, June, 1992.
[16] J. Weng, N. Ahuja and T. S. Huang, "Learning recognition and segmentation of 3-D objects from 2-D images,”
Proc. 4th International Conf. Computer Vision, Berlin,
Germany, pp. 121-128, May, 1993.

26

CHAPTER 2. DEEP LEARNING

[17] J. Weng, N. Ahuja and T. S. Huang, "Learning recognition and segmentation using the Cresceptron,”International
Journal of Computer Vision, vol. 25, no. 2, pp. 105-139,
Nov. 1997.

[33] L. Deng, O. Abdel-Hamid, and D. Yu, A deep convolutional neural network using heterogeneous pooling for trading acoustic invariance with phonetic confusion,
ICASSP, 2013.

[18] Morgan, Bourlard, Renals, Cohen, Franco (1993) “Hybrid neural network/hidden Markov model systems for continuous speech recognition. ICASSP/IJPRAI”

[34] T. Sainath et al., “Convolutional neural networks for
LVCSR,”ICASSP, 2013.

[19] T. Robinson. (1992) A real-time recurrent error propagation network word recognition system, ICASSP.

[35] D. Yu, L. Deng, G. Li, and F. Seide (2011). “Discriminative pretraining of deep neural networks,”U.S. Patent
Filing.

[20] Waibel, Hanazawa, Hinton, Shikano, Lang. (1989)
“Phoneme recognition using time-delay neural networks.
IEEE Transactions on Acoustics, Speech and Signal Processing.”

[36] NIPS Workshop: Deep Learning for Speech Recognition and Related Applications, Whistler, BC, Canada, Dec.
2009 (Organizers: Li Deng, Geoﬀ Hinton, D. Yu).

[21] J. Baker, Li Deng, Jim Glass, S. Khudanpur, C.-H. Lee,
N. Morgan, and D. O'Shaughnessy (2009). “Research
Developments and Directions in Speech Recognition and
Understanding, Part 1,”
IEEE Signal Processing Magazine, vol. 26, no. 3, pp. 75-80, 2009.
[22] Y. Bengio (1991). “Artiﬁcial Neural Networks and their
Application to Speech/Sequence Recognition,”
Ph.D. thesis, McGill University, Canada.
[23] L. Deng, K. Hassanein, M. Elmasry. (1994) “Analysis of correlation structure for a neural predictive model with applications to speech recognition,”
Neural Networks, vol.
7, No. 2., pp. 331-339.

[37] Yu, D.; Deng, L. (2014). “Automatic Speech Recognition: A Deep Learning Approach (Publisher: Springer)".
[38] D. C. Ciresan et al., “Deep Big Simple Neural Nets for
Handwritten Digit Recognition,”
Neural Computation, 22, pp. 3207–3220, 2010.
[39] R. Raina, A. Madhavan, A. Ng., Large-scale Deep Unsu“ pervised Learning using Graphics Processors,”
Proc. 26th
Int. Conf. on Machine Learning, 2009.
[40] Riesenhuber, M; Poggio, T. “Hierarchical models of object recognition in cortex”. Nature Neuroscience 1999
(11): 1019–1025.

[24] Hinton, G.; Deng, L.; Yu, D.; Dahl, G.; Mohamed, [41] Y. LeCun, B. Boser, J. S. Denker, D. Henderson, R. E.
A.; Jaitly, N.; Senior, A.; Vanhoucke, V.; Nguyen,
Howard, W. Hubbard, L. D. Jackel. Backpropagation ApP.; Sainath, T.; Kingsbury, B. (2012). “Deep Neuplied to Handwritten Zip Code Recognition. Neural Comral Networks for Acoustic Modeling in Speech Recogputation, 1(4):541–551, 1989. nition --- The shared views of four research groups”
. IEEE Signal Processing Magazine 29 (6): 82–97. [42] S. Hochreiter. Untersuchungen zu dynamischen neudoi:10.1109/msp.2012.2205597. ronalen Netzen. Diploma thesis, Institut f. Informatik,
Technische Univ. Munich, 1991. Advisor: J. Schmidhu[25] Deng, L.; Hinton, G.; Kingsbury, B. (2013). “New types ber of deep neural network learning for speech recognition and related applications: An overview (ICASSP)".
[43] S. Hochreiter, Y. Bengio, P. Frasconi, and J. Schmid[26] Keynote talk: Recent Developments in Deep Neural Networks. ICASSP, 2013 (by Geoﬀ Hinton).
[27] Keynote talk: “Achievements and Challenges of Deep
Learning - From Speech Analysis and Recognition
To Language and Multimodal Processing,”Interspeech,
September 2014.
[28] G. E. Hinton., “Learning multiple layers of representation,”Trends in Cognitive Sciences, 11, pp. 428–434,
2007.
[29] J. Schmidhuber., Learning complex, extended sequences
“
using the principle of history compression,”Neural Computation, 4, pp. 234–242, 1992.
[30] J. Schmidhuber., “My First Deep Learning System of
1991 + Deep Learning Timeline 1962–2013.”
[31] http://research.microsoft.com/apps/pubs/default.aspx? id=189004 [32] L. Deng et al. Recent Advances in Deep Learning for
Speech Research at Microsoft, ICASSP, 2013.

huber. Gradient ﬂow in recurrent nets: the diﬃculty of learning long-term dependencies. In S. C. Kremer and J.
F. Kolen, editors, A Field Guide to Dynamical Recurrent
Neural Networks. IEEE Press, 2001.
[44] Hochreiter, Sepp; and Schmidhuber, Jürgen; Long ShortTerm Memory, Neural Computation, 9(8):1735–1780,
1997
[45] Graves, Alex; and Schmidhuber, Jürgen; Oﬄine Handwriting Recognition with Multidimensional Recurrent Neural Networks, in Bengio, Yoshua; Schuurmans, Dale; Lafferty, John; Williams, Chris K. I.; and Culotta, Aron
(eds.), Advances in Neural Information Processing Systems
22 (NIPS'22), December 7th–10th, 2009, Vancouver, BC,
Neural Information Processing Systems (NIPS) Foundation, 2009, pp. 545–552
[46] Graves, A.; Liwicki, M.; Fernandez, S.; Bertolami, R.;
Bunke, H.; Schmidhuber, J.“A Novel Connectionist System for Improved Unconstrained Handwriting Recognition”. IEEE Transactions on Pattern Analysis and Machine Intelligence 31 (5): 2009.

2.11. REFERENCES

[47] Sven Behnke (2003). Hierarchical Neural Networks for
Image Interpretation. (PDF). Lecture Notes in Computer
Science 2766. Springer.
[48] Smolensky, P. (1986). Information processing in dynamical systems: Foundations of harmony theory. In D. E.
Rumelhart, J. L. McClelland, & the PDP Research Group,
Parallel Distributed Processing: Explorations in the Microstructure of Cognition. 1. pp. 194–281.
[49] Hinton, G. E.; Osindero, S.; Teh, Y. (2006).
“A fast learning algorithm for deep belief nets”
(PDF). Neural Computation 18 (7):
1527–1554.
doi:10.1162/neco.2006.18.7.1527. PMID 16764513.
[50] Hinton, G. (2009).“Deep belief networks” Scholarpedia
.
4 (5): 5947. doi:10.4249/scholarpedia.5947.
[51] John Markoﬀ (25 June 2012). “How Many Computers to Identify a Cat? 16,000.”. New York Times.
[52] Ng, Andrew; Dean, Jeﬀ (2012). “Building High-level
Features Using Large Scale Unsupervised Learning”
(PDF).
[53] D. C. Ciresan, U. Meier, J. Masci, L. M. Gambardella, J.
Schmidhuber. Flexible, High Performance Convolutional
Neural Networks for Image Classiﬁcation. International
Joint Conference on Artiﬁcial Intelligence (IJCAI-2011,
Barcelona), 2011.
[54] Martines, H.; Bengio, Y.; Yannakakis, G. N. (2013).
“Learning Deep Physiological Models of Aﬀect”. IEEE
Computational Intelligence 8 (2): 20.
[55] D. C. Ciresan, U. Meier, J. Masci, J. Schmidhuber. MultiColumn Deep Neural Network for Traﬃc Sign Classiﬁcation. Neural Networks, 2012.
[56] D. Ciresan, A. Giusti, L. Gambardella, J. Schmidhuber.
Deep Neural Networks Segment Neuronal Membranes in
Electron Microscopy Images. In Advances in Neural Information Processing Systems (NIPS 2012), Lake Tahoe,
2012.
[57] D. C. Ciresan, U. Meier, J. Schmidhuber. Multi-column
Deep Neural Networks for Image Classiﬁcation. IEEE
Conf. on Computer Vision and Pattern Recognition
CVPR 2012.
[58] D. J. Felleman and D. C. Van Essen, "Distributed hierarchical processing in the primate cerebral cortex,”
Cerebral
Cortex, 1, pp. 1-47, 1991.
[59] J. Weng, "Natural and Artiﬁcial Intelligence: Introduction to Computational Brain-Mind,”BMI Press, ISBN 9780985875725, 2012.
[60] J. Weng, "Why Have We Passed `Neural Networks Do not
Abstract Well'?,”
Natural Intelligence: the INNS Magazine, vol. 1, no.1, pp. 13-22, 2011.
[61] Z. Ji, J. Weng, and D. Prokhorov, "Where-What Network
1: Where and What Assist Each Other Through Top-down
Connections,”Proc. 7th International Conference on Development and Learning (ICDL'08), Monterey, CA, Aug.
9-12, pp. 1-6, 2008.

27

[62] X. Wu, G. Guo, and J. Weng, "Skull-closed Autonomous
Development: WWN-7 Dealing with Scales,”Proc. International Conference on Brain-Mind, July 27–28, East
Lansing, Michigan, pp. +1-9, 2013.
[63] Szegedy, Christian, Alexander Toshev, and Dumitru Erhan. “Deep neural networks for object detection.”Advances in Neural Information Processing Systems. 2013.
[64] T. Mikolov et al., “Recurrent neural network based language model,”Interspeech, 2010.
[65] LeCun, Y. et al. “Gradient-based learning applied to document recognition” Proceedings of the IEEE 86 (11):
.
2278–2324. doi:10.1109/5.726791.
[66] G. E. Hinton et al.., “Deep Neural Networks for Acoustic Modeling in Speech Recognition: The shared views of four research groups,”IEEE Signal Processing Magazine, pp. 82–97, November 2012.
[67] Y. Bengio et al..,“Advances in optimizing recurrent networks,”ICASSP, 2013.
[68] G. Dahl et al..,“Improving DNNs for LVCSR using rectiﬁed linear units and dropout,”ICASSP, 2013.
[69] G. E. Hinton.,“A Practical Guide to Training Restricted
Boltzmann Machines,”Tech. Rep. UTML TR 2010-003,
Dept. CS., Univ. of Toronto, 2010.
[70] Hinton, G.E. “Deep belief networks”. Scholarpedia 4
(5): 5947. doi:10.4249/scholarpedia.5947.
[71] H. Larochelle et al.., “An empirical evaluation of deep architectures on problems with many factors of variation,” in Proc. 24th Int. Conf. Machine Learning, pp. 473–480,
2007.
[72] G. E. Hinton.,“Training Product of Experts by Minimizing Contrastive Divergence,”
Neural Computation, 14, pp.
1771–1800, 2002.
[73] A. Fischer and C. Igel. Training Restricted Boltzmann
Machines: An Introduction. Pattern Recognition 47, pp.
25-39, 2014
[74] http://ufldl.stanford.edu/tutorial/index.php/
Convolutional_Neural_Network
[75]
[76]
[77] Hinton, Geoﬀrey; Salakhutdinov, Ruslan (2012).“A better way to pretrain deep Boltzmann machines” (PDF).
Advances in Neural 3: 1–9.
[78] Hinton, Geoﬀrey; Salakhutdinov, Ruslan (2009). “Eﬃcient Learning of Deep Boltzmann Machines” (PDF) 3. pp. 448–455.
[79] Bengio, Yoshua; LeCun, Yann (2007).“Scaling Learning
Algorithms towards AI” (PDF) 1. pp. 1–41.
[80] Larochelle, Hugo; Salakhutdinov, Ruslan (2010). “Efﬁcient Learning of Deep Boltzmann Machines” (PDF). pp. 693–700.

28

CHAPTER 2. DEEP LEARNING

[81] Vincent, Pascal; Larochelle, Hugo; Lajoie, Isabelle; Bengio, Yoshua; Manzagol, Pierre-Antoine (2010).“Stacked
Denoising Autoencoders: Learning Useful Representations in a Deep Network with a Local Denoising Criterion”. The Journal of Machine Learning Research 11:
3371–3408.

[96] Lee, Honglak; Grosse, Roger (2009). “Convolutional deep belief networks for scalable unsupervised learning of hierarchical representations”. Proceedings of the 26th
Annual International Conference on Machine Learning ICML '09: 1–8.

[97] Lin, Yuanqing; Zhang, Tong (2010).“Deep Coding Network” (PDF). Advances in Neural . . .: 1–9.
[82] Deng, Li; Yu, Dong (2011). “Deep Convex Net: A
Scalable Architecture for Speech Pattern Classiﬁcation”
[98] Ranzato, Marc Aurelio; Boureau, Y-Lan (2007).“Sparse
(PDF). Proceedings of the Interspeech: 2285–2288.
Feature Learning for Deep Belief Networks”(PDF). Advances in neural information . . .: 1–8.
[83] Deng, Li; Yu, Dong; Platt, John (2012).“Scalable stacking and learning for building deep architectures”. 2012
[99] Socher, Richard; Lin, Clif (2011). “Parsing Natural
IEEE International Conference on Acoustics, Speech and
Scenes and Natural Language with Recursive Neural NetSignal Processing (ICASSP): 2133–2136. works” (PDF). Proceedings of the . . .
[84] David, Wolpert (1992). “Stacked generalization”.
[100] Taylor, Graham; Hinton, Geoﬀrey (2006). “Modeling
Neural networks 5(2): 241–259. doi:10.1016/S0893Human Motion Using Binary Latent Variables” (PDF).
6080(05)80023-1.
Advances in neural . . .
[85] Bengio, Yoshua (2009).“Learning deep architectures for
[101] Vincent, Pascal; Larochelle, Hugo (2008). “ExtractAI”. Foundations and trends in Machine Learning 2(1): ing and composing robust features with denoising autoen1–127. coders” Proceedings of the 25th international conference
.
on Machine learning - ICML '08: 1096–1103.
[86] Hutchinson, Brian; Deng, Li; Yu, Dong (2012). “Tensor deep stacking networks”. IEEE transactions on pattern
[102] Kemp, Charles; Perfors, Amy; Tenenbaum, Joshua analysis and machine intelligence 1–15.
(2007). “Learning overhypotheses with hierarchical
Bayesian models”. Developmental science. 10(3):
[87] Hinton, Geoﬀrey; Salakhutdinov, Ruslan (2006). “Re307–21. doi:10.1111/j.1467-7687.2007.00585.x. PMID ducing the Dimensionality of Data with Neural Networks”
17444972.
. Science 313: 504–507. doi:10.1126/science.1127647.
PMID 16873662.

[103] Xu, Fei; Tenenbaum, Joshua (2007). “Word learning as Bayesian inference”. Psychol Rev. 114(2): 245–72.
[88] Dahl, G.; Yu, D.; Deng, L.; Acero, A. (2012). “Contextdoi:10.1037/0033-295X.114.2.245. PMID 17500627.
Dependent Pre-Trained Deep Neural Networks for LargeVocabulary Speech Recognition”. Audio, Speech, and ... [104] Chen, Bo; Polatkan, Gungor (2011). “The Hierarchical
20(1): 30–42.
Beta Process for Convolutional Factor Analysis and Deep
Learning” (PDF). Machine Learning . . .
[89] Mohamed, Abdel-rahman; Dahl, George; Hinton, Geoffrey (2012).“Acoustic Modeling Using Deep Belief Net- [105] Fei-Fei, Li; Fergus, Rob (2006). “One-shot learning of works”. IEEE Transactions on Audio, Speech, and Lanobject categories”. IEEE Trans Pattern Anal Mach Intell. guage Processing. 20(1): 14–22.
28(4): 594–611. doi:10.1109/TPAMI.2006.79. PMID
16566508.
[90] Courville, Aaron; Bergstra, James; Bengio, Yoshua
(2011). “A Spike and Slab Restricted Boltzmann Ma- [106] Rodriguez, Abel; Dunson, David (2008).
“The
chine” (PDF). International . . . 15: 233–241.
Nested Dirichlet Process”. Journal of the American Statistical Association. 103(483): 1131–1154.
[91] Mitchell, T; Beauchamp, J (1988). “Bayesian Varidoi:10.1198/016214508000000553. able Selection in Linear Regression”. Journal of the
American Statistical Association. 83 (404): 1023–1032. [107] Ruslan, Salakhutdinov; Joshua, Tenenbaum (2012). doi:10.1080/01621459.1988.10478694. “Learning with Hierarchical-Deep Models”. IEEE transactions on pattern analysis and machine intelligence: 1–14.
[92] Courville, Aaron; Bergstra, James; Bengio, Yoshua
PMID 23267196.
(2011). “Unsupervised Models of Images by Spike-andSlab RBMs” (PDF). Proceedings of the . . . 10: 1–8.
[108] Chalasani, Rakesh; Principe, Jose (2013). “Deep Predictive Coding Networks”. arXiv preprint arXiv: 1–13.
[93] Hinton, Geoﬀrey; Osindero, Simon; Teh, Yee-Whye
(2006). “A Fast Learning Algorithm for Deep Belief [109] Cho, Youngmin (2012). “Kernel Methods for Deep
Nets”. Neural Computation 1554: 1527–1554.
Learning” (PDF). pp. 1–9.
[94] Larochelle, Hugo; Bengio, Yoshua; Louradour, Jerdme; [110] Scholkopf, B; Smola, Alexander (1998). “Nonlinear
Lamblin, Pascal (2009).“Exploring Strategies for Traincomponent analysis as a kernel eigenvalue problem”. ing Deep Neural Networks”. The Journal of Machine
Neural computation (44).
Learning Research 10: 1–40.
[111] L. Deng, G. Tur, X. He, and D. Hakkani-Tur. “Use of
[95] Coates, Adam; Carpenter, Blake (2011). “Text DetecKernel Deep Convex Networks and End-To-End Learning tion and Character Recognition in Scene Images with Unfor Spoken Language Understanding,”Proc. IEEE Worksupervised Feature Learning”. pp. 440–445. shop on Spoken Language Technologies, 2012

2.11. REFERENCES

29

[112] Mnih, Volodymyr et al. (2015). “Human-level control [131] Y. Bengio, R. Ducharme, P. Vincent, C. Jauvin.,“A Neuthrough deep reinforcement learning” (PDF) 518. pp. ral Probabilistic Language Model,”Journal of Machine
529–533.
Learning Research 3 (2003) 1137–1155', 2003.
[113] TIMIT Acoustic-Phonetic Continuous Speech Corpus Lin- [132] Goldberg, Yoav; Levy, Omar. “word2vec Explained: guistic Data Consortium, Philadelphia.
Deriving Mikolov et al.’s Negative-Sampling WordEmbedding Method” (PDF). Arxiv. Retrieved 26 Oc[114] Abdel-Hamid, O. et al. (2014). “Convolutional Neural tober 2014.
Networks for Speech Recognition” IEEE/ACM Transac. tions on Audio, Speech, and Language Processing 22 (10): [133] Socher, Richard; Manning, Christopher.“Deep Learning
1533–1545. doi:10.1109/taslp.2014.2339736. for NLP” (PDF). Retrieved 26 October 2014.
[115] Deng, L.; Platt, J. (2014). “Ensemble Deep Learning for [134] Socher, Richard; Bauer, John; Manning, Christopher; Ng,
Speech Recognition”. Proc. Interspeech.
Andrew (2013). “Parsing With Compositional Vector
Grammars”(PDF). Proceedings of the ACL 2013 confer[116] Yu, D.; Deng, L. (2010). “Roles of Pre-Training ence. and Fine-Tuning in Context-Dependent DBN-HMMs for
Real-World Speech Recognition”. NIPS Workshop on [135] Socher, Richard (2013). “Recursive Deep Models for
Deep Learning and Unsupervised Feature Learning.
Semantic Compositionality Over a Sentiment Treebank”
(PDF). EMNLP 2013.
[117] Deng L., Li, J., Huang, J., Yao, K., Yu, D., Seide, F. et al.
Recent Advances in Deep Learning for Speech Research [136] Y. Shen, X. He, J. Gao, L. Deng, and G. Mesnil (2014) at Microsoft. ICASSP, 2013.
" A Latent Semantic Model with Convolutional-Pooling
Structure for Information Retrieval,”Proc. CIKM.
[118] Deng, L.; Li, Xiao (2013).“Machine Learning Paradigms for Speech Recognition: An Overview”. IEEE Transac[137] P. Huang, X. He, J. Gao, L. Deng, A. Acero, and L. Heck tions on Audio, Speech, and Language Processing.
(2013) “Learning Deep Structured Semantic Models for
Web Search using Clickthrough Data,”Proc. CIKM.
[119] L. Deng, M. Seltzer, D. Yu, A. Acero, A. Mohamed, and
G. Hinton (2010) Binary Coding of Speech Spectrograms
[138] I. Sutskever, O. Vinyals, Q. Le (2014) “Sequence to SeUsing a Deep Auto-encoder. Interspeech. quence Learning with Neural Networks,”Proc. NIPS.
[120] Z. Tuske, P. Golik, R. Schlüter and H. Ney (2014).
Acoustic Modeling with Deep Neural Networks Using [139] J. Gao, X. He, W. Yih, and L. Deng(2014) “Learning
Continuous Phrase Representations for Translation ModRaw Time Signal for LVCSR. Interspeech. eling,”Proc. ACL.
[121] McMillan, R.“How Skype Used AI to Build Its Amazing
[140] J. Gao, P. Pantel, M. Gamon, X. He, L. Deng (2014)
New Language Translator”, Wire, Dec. 2014.
“Modeling Interestingness with Deep Neural Networks,”
Proc. EMNLP.
[122] Hannun et al. (2014) “Deep Speech: Scaling up end-toend speech recognition”, arXiv:1412.5567.
[141] J. Gao, X. He, L. Deng (2014) “Deep Learning for Nat[123] Ron Schneiderman (2015) “Accuracy, Apps Advance ural Language Processing: Theory and Practice (TutoSpeech Recognition --- Interview with Vlad Sejnoha and rial),”CIKM. Li Deng”, IEEE Signal Processing Magazine, Jan, 2015.
[142] Arrowsmith, J; Miller, P (2013). “Trial watch: Phase
[124] http://yann.lecun.com/exdb/mnist/.
II and phase III attrition rates 2011-2012”. Nature Reviews Drug Discovery 12 (8): 569. doi:10.1038/nrd4090.
[125] D. Ciresan, U. Meier, J. Schmidhuber., “Multi-column
PMID 23903212.
Deep Neural Networks for Image Classiﬁcation,”
Technical Report No. IDSIA-04-12', 2012.
[143] Verbist, B; Klambauer, G; Vervoort, L; Talloen, W;
The Qstar, Consortium; Shkedy, Z; Thas, O; Ben[126] Vinyals et al. (2014)."Show and Tell: A Neural Image der, A; Göhlmann, H. W.; Hochreiter, S (2015).
Caption Generator,”arXiv:1411.4555.
“Using transcriptomics to guide lead optimization in drug discovery projects: Lessons learned from
[127] Fang et al. (2014)."From Captions to Visual Concepts and the QSTAR project”.
Drug Discovery Today.
Back,”arXiv:1411.4952.
doi:10.1016/j.drudis.2014.12.014. PMID 25582842.
[128] Kiros et al. (2014)."Unifying Visual-Semantic Embeddings with Multimodal Neural Language Models,” arXiv: [144]“Announcement of the winners of the Merck Molecular Activity Challenge”https://www.kaggle.com/c/
1411.2539.
MerckActivity/details/winners.
[129] Zhong, S.; Liu, Y.; Liu, Y. “Bilinear Deep Learning for
Image Classiﬁcation”. Proceedings of the 19th ACM In- [145] Dahl, G. E.; Jaitly, N.; & Salakhutdinov, R. (2014)
“Multi-task Neural Networks for QSAR Predictions,” ternational Conference on Multimedia 11: 343–352.
ArXiv, 2014.
[130] Nvidia Demos a Car Computer Trained with “Deep
Learning” (2015-01-06), David Talbot, MIT Technology [146]“Toxicology in the 21st century Data Challenge”https:
Review
//tripod.nih.gov/tox21/challenge/leaderboard.jsp

30

CHAPTER 2. DEEP LEARNING

[147]“NCATS Announces Tox21 Data Challenge Winners” [162] http://www.ncats.nih.gov/news-and-events/features/ tox21-challenge-winners.html
[163]
[148] Unterthiner, T.; Mayr, A.; Klambauer, G.; Steijaert, M.;
Ceulemans, H.; Wegner, J. K.; & Hochreiter, S. (2014)
“Deep Learning as an Opportunity in Virtual Screening”
. Workshop on Deep Learning and Representation Learn- [164] ing (NIPS2014).

Smith, G. W. (March 27, 2015). “Art and Artiﬁcial Intelligence”. ArtEnt. Retrieved March 27, 2015.
Ben Goertzel. Are there Deep Reasons Underlying the
Pathologies of Today’s Deep Learning Algorithms?
(2015) Url: http://goertzel.org/DeepLearning_v1.pdf
Nguyen, Anh, Jason Yosinski, and Jeﬀ Clune. “Deep
Neural Networks are Easily Fooled: High Conﬁdence
Predictions for Unrecognizable Images.”arXiv preprint arXiv:1412.1897 (2014).

[149] Unterthiner, T.; Mayr, A.; Klambauer, G.; & Hochreiter,
S. (2015) “Toxicity Prediction using Deep Learning”.
[165] Szegedy, Christian, et al.“Intriguing properties of neural
ArXiv, 2015. networks.”arXiv preprint arXiv:1312.6199 (2013).
[150] Ramsundar, B.; Kearnes, S.; Riley, P.; Webster, D.; Kon[166] Zhu, S.C.; Mumford, D. “A stochastic grammar of imerding, D.;& Pande, V. (2015)“Massively Multitask Netages”. Found. Trends. Comput. Graph. Vis. 2 (4): works for Drug Discovery”. ArXiv, 2015.
259–362. doi:10.1561/0600000018.

[151] Tkachenko, Yegor. Autonomous CRM Control via CLV [167] Miller, G. A., and N. Chomsky. “Pattern conception.”
Approximation with Deep Reinforcement Learning in
Paper for Conference on pattern detection, University of
Discrete and Continuous Action Space. (April 8, 2015).
Michigan. 1957. arXiv.org: http://arxiv.org/abs/1504.01840
[168] Jason Eisner, Deep Learning of Recursive Struc[152] Utgoﬀ, P. E.; Stracuzzi, D. J. (2002). “Manyture:
Grammar Induction, http://techtalks.tv/talks/ layered learning”. Neural Computation 14: 2497–2529. deep-learning-of-recursive-structure-grammar-induction/ doi:10.1162/08997660260293319.
58089/
[153] J. Elman, et al., “Rethinking Innateness,”1996.
[154] Shrager, J.; Johnson, MH (1996). “Dynamic plasticity inﬂuences the emergence of function in a simple cortical array”. Neural Networks 9 (7): 1119–1129. doi:10.1016/0893-6080(96)00033-0. [155] Quartz, SR; Sejnowski, TJ (1997). “The neural basis of cognitive development: A constructivist manifesto”. Behavioral and Brain Sciences 20 (4): 537–556. doi:10.1017/s0140525x97001581. [156] S. Blakeslee., “In brain's early growth, timetable may be critical,”The New York Times, Science Section, pp. B5–
B6, 1995.
[157] {BUFILL} E. Buﬁll, J. Agusti, R. Blesa., “Human neoteny revisited: The case of synaptic plasticity,”American Journal of Human Biology, 23 (6), pp. 729–739,
2011.
[158] J. Shrager and M. H. Johnson., “Timing in the development of cortical function: A computational approach,”In
B. Julesz and I. Kovacs (Eds.), Maturational windows and adult cortical plasticity, 1995.
[159] D. Hernandez., “The Man Behind the Google
Brain: Andrew Ng and the Quest for the New
AI,”http://www.wired.com/wiredenterprise/2013/
Wired, 10 May
05/neuro-artificial-intelligence/all/.
2013.
[160] C. Metz., “Facebook's 'Deep Learning' Guru Reveals http://www.wired.com/wiredenterprise/ the Future of AI,”
2013/12/facebook-yann-lecun-qa/. Wired, 12 December
2013.
[161] G. Marcus., Is Deep Learning” Revolution in Artiﬁcial
““
a
Intelligence?" The New Yorker, 25 November 2012.

2.12 External links
• TED talk on the applications of deep learning and future consequences by Jeremy Howard
• Deep learning information from the University of
Montreal
• Deep learning information from Stanford University
• Deep Learning Resources, NVIDIA Developer
Zone
• Geoﬀrey Hinton's webpage
• Hinton deep learning tutorial
• Yann LeCun's webpage
• The Center for Biological and Computational
Learning (CBCL)
• Stanford tutorial on unsupervised feature learning and deep learning
• Google's DistBelief Framework
• NIPS 2013 Conference (talks on deep learning related material)
• Mnih, Volodymyr; Kavukcuoglu, Koray; Silver,
David; Graves, Alex; Antonoglou, Ioannis; Wierstra, Daan; Riedmiller, Martin (2013), Playing
Atari with Deep Reinforcement Learning (PDF), arXiv:1312.5602 2.12. EXTERNAL LINKS
• 100 Best GitHub: Deep Learning
• Silicon Chips That See Are Going to Make Your
Smartphone Brilliant.“Many of the devices around us may soon acquire powerful new abilities to understand images and video, thanks to hardware designed for the machine-learning technique called deep learning.”Tom Simonite (May 2015), MIT
Technology Review

31

Chapter 3

Feature learning
Feature learning or representation learning* [1] is a set of techniques that learn a transformation of raw data input to a representation that can be eﬀectively exploited in machine learning tasks.
Feature learning is motivated by the fact that machine learning tasks such as classiﬁcation often require input that is mathematically and computationally convenient to process. However, real-world data such as images, video, and sensor measurement is usually complex, redundant, and highly variable. Thus, it is necessary to discover useful features or representations from raw data. Traditional hand-crafted features often require expensive human labor and often rely on expert knowledge. Also, they normally do not generalize well. This motivates the design of eﬃcient feature learning techniques.

weights may be found by minimizing the average representation error (over the input data), together with a L1 regularization on the weights to enable sparsity (i.e., the representation of each data point has only a few nonzero weights). Supervised dictionary learning exploits both the structure underlying the input data and the labels for optimizing the dictionary elements. For example, a supervised dictionary learning technique was proposed by Mairal et al. in
2009.* [6] The authors apply dictionary learning on classiﬁcation problems by jointly optimizing the dictionary elements, weights for representing data points, and parameters of the classiﬁer based on the input data. In particular, a minimization problem is formulated, where the objective function consists of the classiﬁcation error, the repreFeature learning can be divided into two categories: su- sentation error, an L1 regularization on the representing weights for each data point (to enable sparse representapervised and unsupervised feature learning. tion of data), and an L2 regularization on the parameters of the classiﬁer.
• In supervised feature learning, features are learned with labeled input data. Examples include neural networks, multilayer perceptron, and (supervised) dictionary learning.
• In unsupervised feature learning, features are
3.1.2 Neural networks learned with unlabeled input data. Examples include dictionary learning, independent component analysis, autoencoders, matrix factorization,* [2] and var- Neural networks are used to illustrate a family of learnious forms of clustering.* [3]* [4]* [5] ing algorithms via a “network”consisting of multiple layers of inter-connected nodes. It is inspired by the nervous system, where the nodes are viewed as neurons and edges are viewed as synapse. Each edge has an as3.1 Supervised feature learning sociated weight, and the network deﬁnes computational rules that passes input data from the input layer to the
Supervised feature learning is to learn features from laoutput layer. A network function associated with a neubeled data. Several approaches are introduced in the folral network characterizes the relationship between input lowing. and output layers, which is parameterized by the weights.
With appropriately deﬁned network functions, various learning tasks can be performed by minimizing a cost
3.1.1 Supervised dictionary learning function over the network function (weights).
Dictionary learning is to learn a set (dictionary) of representative elements from the input data such that each data point can be represented as a weighted sum of the representative elements. The dictionary elements and the

Multilayer neural networks can be used to perform feature learning, since they learn a representation of their input at the hidden layer(s) which is subsequently used for classiﬁcation or regression at the output layer.

32

3.2. UNSUPERVISED FEATURE LEARNING

3.2 Unsupervised feature learning
Unsupervised feature learning is to learn features from unlabeled data. The goal of unsupervised feature learning is often to discover low-dimensional features that captures some structure underlying the high-dimensional input data. When the feature learning is performed in an unsupervised way, it enables a form of semisupervised learning where ﬁrst, features are learned from an unlabeled dataset, which are then employed to improve performance in a supervised setting with labeled data.* [7]* [8] Several approaches are introduced in the following. 3.2.1

K-means clustering

K-means clustering is an approach for vector quantization. In particular, given a set of n vectors, k-means clustering groups them into k clusters (i.e., subsets) in such a way that each vector belongs to the cluster with the closest mean. The problem is computationally NP-hard, and suboptimal greedy algorithms have been developed for kmeans clustering.
In feature learning, k-means clustering can be used to group an unlabeled set of inputs into k clusters, and then use the centroids of these clusters to produce features.
These features can be produced in several ways. The simplest way is to add k binary features to each sample, where each feature j has value one iﬀ the jth centroid learned by k-means is the closest to the sample under consideration.* [3] It is also possible to use the distances to the clusters as features, perhaps after transforming them through a radial basis function (a technique that has used to train
RBF networks* [9]). Coates and Ng note that certain variants of k-means behave similarly to sparse coding algorithms.* [10]
In a comparative evaluation of unsupervised feature learning methods, Coates, Lee and Ng found that kmeans clustering with an appropriate transformation outperforms the more recently invented auto-encoders and
RBMs on an image classiﬁcation task.* [3] K-means has also been shown to improve performance in the domain of NLP, speciﬁcally for named-entity recognition;* [11] there, it competes with Brown clustering, as well as with distributed word representations (also known as neural word embeddings).* [8]

3.2.2

Principal component analysis

Principal component analysis (PCA) is often used for dimension reduction. Given a unlabeled set of n input data vectors, PCA generates p (which is much smaller than the dimension of the input data) right singular vectors corresponding to the p largest singular values of the data matrix, where the kth row of the data matrix is the kth input data vector shifted by the sample mean of the input

33
(i.e., subtracting the sample mean from the data vector).
Equivalently, these singular vectors are the eigenvectors corresponding to the p largest eigenvalues of the sample covariance matrix of the input vectors. These p singular vectors are the feature vectors learned from the input data, and they represent directions along which the data has the largest variations.
PCA is a linear feature learning approach since the p singular vectors are linear functions of the data matrix. The singular vectors can be generated via a simple algorithm with p iterations. In the ith iteration, the projection of the data matrix on the (i-1)th eigenvector is subtracted, and the ith singular vector is found as the right singular vector corresponding to the largest singular of the residual data matrix. PCA has several limitations. First, it assumes that the directions with large variance are of most interest, which may not be the case in many applications. PCA only relies on orthogonal transformations of the original data, and it only exploits the ﬁrst- and second-order moments of the data, which may not well characterize the distribution of the data. Furthermore, PCA can eﬀectively reduce dimension only when the input data vectors are correlated
(which results in a few dominant eigenvalues).

3.2.3 Local linear embedding
Local linear embedding (LLE) is a nonlinear unsupervised learning approach for generating low-dimensional neighbor-preserving representations from (unlabeled) high-dimension input. The approach was proposed by
Sam T. Roweis and Lawrence K. Saul in 2000.* [12]* [13]
The general idea of LLE is to reconstruct the original high-dimensional data using lower-dimensional points while maintaining some geometric properties of the neighborhoods in the original data set. LLE consists of two major steps. The ﬁrst step is for “neighborpreserving,”where each input data point Xi is reconstructed as a weighted sum of K nearest neighboring data points, and the optimal weights are found by minimizing the average squared reconstruction error (i.e., diﬀerence between a point and its reconstruction) under the constraint that the weights associated to each point sum up to one. The second step is for “dimension reduction,” by looking for vectors in a lower-dimensional space that minimizes the representation error using the optimized weights in the ﬁrst step. Note that in the ﬁrst step, the weights are optimized with data being ﬁxed, which can be solved as a least squares problem; while in the second step, lower-dimensional points are optimized with the weights being ﬁxed, which can be solved via sparse eigenvalue decomposition.
The reconstruction weights obtained in the ﬁrst step captures the“intrinsic geometric properties”of a neighborhood in the input data.* [13] It is assumed that original data lie on a smooth lower-dimensional manifold, and the

34

CHAPTER 3. FEATURE LEARNING

“intrinsic geometric properties”captured by the weights of the original data are expected also on the manifold.
This is why the same weights are used in the second step of LLE. Compared with PCA, LLE is more powerful in exploiting the underlying structure of data.

den variables, a group of visible variables, and edges connecting the hidden and visible nodes. It is a special case of the more general Boltzmann machines with the constraint of no intra-node connections. Each edge in an RBM is associated with a weight. The weights together with the connections deﬁne an energy function, based on which a joint distribution of visible and hidden nodes can be
3.2.4 Independent component analysis devised. Based on the topology of the RBM, the hidden
(visible) variables are independent conditioned on the visIndependent component analysis (ICA) is technique for ible (hidden) variables. Such conditional independence learning a representation of data using a weighted sum facilitates computations on RBM. of independent non-Gaussian components.* [14] The asAn RBM can be viewed as a single layer architecture for sumption of non-Gaussian is imposed since the weights unsupervised feature learning. In particular, the visible cannot be uniquely determined when all the components variables correspond to input data, and the hidden varifollow Gaussian distribution. ables correspond to feature detectors. The weights can be trained by maximizing the probability of visible variables using the contrastive divergence (CD) algorithm by
3.2.5 Unsupervised dictionary learning
Geoﬀrey Hinton.* [18]
Diﬀerent from supervised dictionary learning, unsupervised dictionary learning does not utilize the labels of the data and only exploits the structure underlying the data for optimizing the dictionary elements. An example of unsupervised dictionary learning is sparse coding, which aims to learn basis functions (dictionary elements) for data representation from unlabeled input data. Sparse coding can be applied to learn overcomplete dictionary, where the number of dictionary elements is larger than the dimension of the input data.* [15] Aharon et al. proposed an algorithm known as K-SVD for learning from unlabeled input data a dictionary of elements that enables sparse representation of the data.* [16]

3.3 Multilayer/Deep architectures
The hierarchical architecture of the neural system inspires deep learning architectures for feature learning by stacking multiple layers of simple learning blocks.* [17]
These architectures are often designed based on the assumption of distributed representation: observed data is generated by the interactions of many diﬀerent factors on multiple levels. In a deep learning architecture, the output of each intermediate layer can be viewed as a representation of the original input data. Each level uses the representation produced by previous level as input, and produces new representations as output, which is then fed to higher levels. The input of bottom layer is the raw data, and the output of the ﬁnal layer is the ﬁnal lowdimensional feature or representation.

In general, the training of RBM by solving the above maximization problem tends to result in non-sparse representations. The sparse RBM, * [19] a modiﬁcation of the RBM, was proposed to enable sparse representations.
The idea is to add a regularization term in the objective function of data likelihood, which penalizes the deviation of the expected hidden variables from a small constant p
.

3.3.2 Autoencoder
An autoencoder consisting of encoder and decoder is a paradigm for deep learning architectures. An example is provided by Hinton and Salakhutdinov* [18] where the encoder uses raw data (e.g., image) as input and produces feature or representation as output, and the decoder uses the extracted feature from the encoder as input and reconstructs the original input raw data as output. The encoder and decoder are constructed by stacking multiple layers of RBMs. The parameters involved in the architecture are trained in a greedy layer-by-layer manner: after one layer of feature detectors is learned, they are fed to upper layers as visible variables for training the corresponding
RBM. The process can be repeated until some stopping criteria is satisﬁed.

3.4 See also
• Basis function
• Deep learning

3.3.1

Restricted Boltzmann machine

Restricted Boltzmann machines (RBMs) are often used as a building block for multilayer learning architectures.* [3]* [18] An RBM can be represented by an undirected bipartite graph consisting of a group of binary hid-

• Feature detection (computer vision)
• Feature extraction
• Kernel trick
• Vector quantization

3.5. REFERENCES

3.5 References
[1] Y. Bengio; A. Courville; P. Vincent (2013). “Representation Learning: A Review and New Perspectives” IEEE
.
Trans. PAMI, special issue Learning Deep Architectures.
[2] Nathan Srebro; Jason D. M. Rennie; Tommi S. Jaakkola
(2004). Maximum-Margin Matrix Factorization. NIPS.
[3] Coates, Adam; Lee, Honglak; Ng, Andrew Y. (2011). An analysis of single-layer networks in unsupervised feature learning (PDF). Int'l Conf. on AI and Statistics (AISTATS).
[4] Csurka, Gabriella; Dance, Christopher C.; Fan, Lixin;
Willamowski, Jutta; Bray, Cédric (2004). Visual categorization with bags of keypoints (PDF). ECCV Workshop on Statistical Learning in Computer Vision.
[5] Daniel Jurafsky; James H. Martin (2009). Speech and
Language Processing. Pearson Education International. pp. 145–146.
[6] Mairal, Julien; Bach, Francis; Ponce, Jean; Sapiro,
Guillermo; Zisserman, Andrew (2009).“Supervised Dictionary Learning”. Advances in neural information processing systems.
[7] Percy Liang (2005). Semi-Supervised Learning for Natural Language (PDF) (M. Eng.). MIT. pp. 44–52.
[8] Joseph Turian; Lev Ratinov; Yoshua Bengio (2010).
Word representations: a simple and general method for semi-supervised learning (PDF). Proceedings of the 48th
Annual Meeting of the Association for Computational
Linguistics.
[9] Schwenker, Friedhelm; Kestler, Hans A.; Palm, Günther (2001). “Three learning phases for radial-basisfunction networks”.
Neural Networks 14: 439–
458. doi:10.1016/s0893-6080(01)00027-2. CiteSeerX:
10.1.1.109.312.
[10] Coates, Adam; Ng, Andrew Y. (2012). “Learning feature representations with k-means”. In G. Montavon, G.
B. Orr and K.-R. Müller. Neural Networks: Tricks of the
Trade. Springer.
[11] Dekang Lin; Xiaoyun Wu (2009). Phrase clustering for discriminative learning (PDF). Proc. J. Conf. of the ACL and 4th Int'l J. Conf. on Natural Language Processing of the AFNLP. pp. 1030–1038.
[12] Roweis, Sam T; Saul, Lawrence K (2000). “Nonlinear Dimensionality Reduction by Locally Linear Embedding”. Science, New Series 290 (5500): 2323–2326. doi:10.1126/science.290.5500.2323. [13] Saul, Lawrence K; Roweis, Sam T (2000). “An Introduction to Locally Linear Embedding”.
[14] Hyvärinen, Aapo; Oja, Erkki (2000).“Independent Component Analysis: Algorithms and Applications”. Neural networks (4): 411–430.
[15] Lee, Honglak; Battle, Alexis; Raina, Rajat; Ng, Andrew
Y (2007). “Eﬃcient sparse coding algorithms”. Advances in neural information processing systems.

35

[16] Aharon, Michal; Elad, Michael; Bruckstein, Alfred
(2006).
“K-SVD: An Algorithm for Designing
Overcomplete Dictionaries for Sparse Representation”
. IEEE Trans. Signal Process. 54 (11): 4311–4322. doi:10.1109/TSP.2006.881199. [17] Bengio, Yoshua (2009). “Learning Deep Architectures for AI”. Foundations and Trends® in Machine Learning
2 (1): 1–127. doi:10.1561/2200000006.
[18] Hinton, G. E.; Salakhutdinov, R. R. (2006).
“Reducing the Dimensionality of Data with Neural
Networks” (PDF). Science 313 (5786): 504–507. doi:10.1126/science.1127647. PMID 16873662.
[19] Lee, Honglak; Ekanadham, Chaitanya; Andrew, Ng
(2008). “Sparse deep belief net model for visual area
V2”. Advances in neural information processing systems.

Chapter 4

Unsupervised learning
In machine learning, the problem of unsupervised learning is that of trying to ﬁnd hidden structure in unlabeled data. Since the examples given to the learner are unlabeled, there is no error or reward signal to evaluate a potential solution. This distinguishes unsupervised learning from supervised learning and reinforcement learning.
Unsupervised learning is closely related to the problem of density estimation in statistics.* [1] However unsupervised learning also encompasses many other techniques that seek to summarize and explain key features of the data. Many methods employed in unsupervised learning are based on data mining methods used to preprocess data. Approaches to unsupervised learning include:

4.1 Method of moments
One of the approaches in unsupervised learning is the method of moments. In the method of moments, the unknown parameters (of interest) in the model are related to the moments of one or more random variables, and thus, these unknown parameters can be estimated given the moments. The moments are usually estimated from samples in an empirical way. The basic moments are ﬁrst and second order moments. For a random vector, the ﬁrst order moment is the mean vector, and the second order moment is the covariance matrix (when the mean is zero). Higher order moments are usually represented using tensors which are the generalization of matrices to higher orders as multi-dimensional arrays.

In particular, the method of moments is shown to be eﬀective in learning the parameters of latent variable models.* [5] Latent variable models are statistical models where in addition to the observed variables, a set of la• Approaches for learning latent variable models such tent variables also exists which is not observed. A highly as practical example of latent variable models in machine learning is the topic modeling which is a statistical model
• Expectation–maximization algorithm (EM) for generating the words (observed variables) in the docu• Method of moments ment based on the topic (latent variable) of the document.
In the topic modeling, the words in the document are gen• Blind signal separation techniques, e.g., erated according to diﬀerent statistical parameters when
• Principal component analysis, the topic of the document is changed. It is shown that
• Independent component analysis, method of moments (tensor decomposition techniques) consistently recover the parameters of a large class of la• Non-negative matrix factorization,
*
tent variable models under some assumptions.* [5]
• Singular value decomposition. [3]
Expectation–maximization algorithm (EM) is also one of
Among neural network models, the self-organizing map the most practical methods for learning latent variable
(SOM) and adaptive resonance theory (ART) are com- models. But, it can be stuck in local optima, and the monly used unsupervised learning algorithms. The SOM global convergence of the algorithm to the true unknown is a topographic organization in which nearby locations parameters of the model is not guaranteed. While, for in the map represent inputs with similar properties. The the method of moments, the global convergence is guar*
ART model allows the number of clusters to vary with anteed under some conditions. [5] problem size and lets the user control the degree of similarity between members of the same clusters by means of a user-deﬁned constant called the vigilance parameter. 4.2 See also
ART networks are also used for many pattern recognition tasks, such as automatic target recognition and seismic
• Cluster analysis signal processing. The ﬁrst version of ART was“ART1”
• Expectation–maximization algorithm
, developed by Carpenter and Grossberg (1988).* [4]
• clustering (e.g., k-means, hierarchical clustering),* [2]

mixture

models,

36

4.4. FURTHER READING
• Generative topographic map
• Multilinear subspace learning
• Multivariate analysis
• Radial basis function network

4.3 Notes
[1] Jordan, Michael I.; Bishop, Christopher M. (2004).“Neural Networks”. In Allen B. Tucker. Computer Science
Handbook, Second Edition (Section VII: Intelligent Systems). Boca Raton, FL: Chapman & Hall/CRC Press
LLC. ISBN 1-58488-360-X.
[2] Hastie,Trevor,Robert Tibshirani,
Friedman,Jerome
(2009). The Elements of Statistical Learning: Data mining,Inference,and Prediction. New York: Springer. pp. 485–586. ISBN 978-0-387-84857-0.
[3] Acharyya, Ranjan (2008); A New Approach for Blind
Source Separation of Convolutive Sources, ISBN 978-3639-07797-1 (this book focuses on unsupervised learning with Blind Source Separation)
[4] Carpenter, G.A. and Grossberg, S. (1988). “The ART of adaptive pattern recognition by a self-organizing neural network” (PDF). Computer 21:
77–88.
doi:10.1109/2.33.
[5] Anandkumar, Animashree; Ge, Rong; Hsu, Daniel;
Kakade, Sham; Telgarsky, Matus (2014). “Tensor
Decompositions for Learning Latent Variable Models”
(PDF). Journal of Machine Learning Research (JMLR) 15:
2773−2832.

4.4 Further reading
• Bousquet, O.; von Luxburg, U.; Raetsch, G., eds.
(2004). Advanced Lectures on Machine Learning.
Springer-Verlag. ISBN 978-3540231226.
• Duda, Richard O.; Hart, Peter E.; Stork, David G.
(2001). “Unsupervised Learning and Clustering”
. Pattern classiﬁcation (2nd ed.). Wiley. ISBN 0471-05669-3.
• Hastie, Trevor; Tibshirani, Robert (2009). The
Elements of Statistical Learning: Data mining,Inference,and Prediction. New York: Springer. pp. 485–586. doi:10.1007/978-0-387-848587_14. ISBN 978-0-387-84857-0.
• Hinton, Geoﬀrey; Sejnowski, Terrence J., eds.
(1999). Unsupervised Learning: Foundations of
Neural Computation. MIT Press. ISBN 0-26258168-X. (This book focuses on unsupervised learning in neural networks)

37

Chapter 5

Generative model
In probability and statistics, a generative model is a model for randomly generating observable-data values, typically given some hidden parameters. It speciﬁes a joint probability distribution over observation and label sequences. Generative models are used in machine learning for either modeling data directly (i.e., modeling observations drawn from a probability density function), or as an intermediate step to forming a conditional probability density function. A conditional distribution can be formed from a generative model through Bayes' rule.
Shannon (1948) gives an example in which a table of frequencies of English word pairs is used to generate a sentence beginning with “representing and speedily is an good"; which is not proper English but which will increasingly approximate it as the table is moved from word pairs to word triplets etc.
Generative models contrast with discriminative models, in that a generative model is a full probabilistic model of all variables, whereas a discriminative model provides a model only for the target variable(s) conditional on the observed variables. Thus a generative model can be used, for example, to simulate (i.e. generate) values of any variable in the model, whereas a discriminative model allows only sampling of the target variables conditional on the observed quantities. Despite the fact that discriminative models do not need to model the distribution of the observed variables, they cannot generally express more complex relationships between the observed and target variables. They don't necessarily perform better than generative models at classiﬁcation and regression tasks. In modern applications the two classes are seen as complementary or as diﬀerent views of the same procedure.* [1]
Examples of generative models include:

• Restricted Boltzmann machine
If the observed data are truly sampled from the generative model, then ﬁtting the parameters of the generative model to maximize the data likelihood is a common method.
However, since most statistical models are only approximations to the true distribution, if the model's application is to infer about a subset of variables conditional on known values of others, then it can be argued that the approximation makes more assumptions than are necessary to solve the problem at hand. In such cases, it can be more accurate to model the conditional density functions directly using a discriminative model (see above), although application-speciﬁc details will ultimately dictate which approach is most suitable in any particular case.

5.1 See also
• Discriminative model
• Graphical model

5.2 References
[1] C. M. Bishop and J. Lasserre, Generative or Discriminative? getting the best of both worlds. In Bayesian Statistics
8, Bernardo, J. M. et al. (Eds), Oxford University Press.
3–23, 2007.

5.3 Sources

• Gaussian mixture model and other types of mixture model • Hidden Markov model
• Probabilistic context-free grammar
• Naive Bayes
• Averaged one-dependence estimators
• Latent Dirichlet allocation
38

• Shannon, C.E. (1948) "A Mathematical Theory of
Communication", Bell System Technical Journal, vol. 27, pp. 379–423, 623–656, July, October,
1948

Chapter 6

Neural coding
Neural coding is a neuroscience-related ﬁeld concerned with characterizing the relationship between the stimulus and the individual or ensemble neuronal responses and the relationship among the electrical activity of the neurons in the ensemble.* [1] Based on the theory that sensory and other information is represented in the brain by networks of neurons, it is thought that neurons can encode both digital and analog information.* [2]

recalled in the hippocampus, a brain region known to be central for memory formation.* [5]* [6]* [7] Neuroscientists have initiated several large-scale brain decoding projects.* [8]* [9]

6.1 Overview

The link between stimulus and response can be studied from two opposite points of view. Neural encoding refers to the map from stimulus to response. The main focus is to understand how neurons respond to a wide variety of stimuli, and to construct models that attempt to predict responses to other stimuli. Neural decoding refers to the reverse map, from response to stimulus, and the challenge is to reconstruct a stimulus, or certain aspects of that stimulus, from the spike sequences it evokes.

Neurons are remarkable among the cells of the body in their ability to propagate signals rapidly over large distances. They do this by generating characteristic electrical pulses called action potentials: voltage spikes that can travel down nerve ﬁbers. Sensory neurons change their activities by ﬁring sequences of action potentials in various temporal patterns, with the presence of external sensory stimuli, such as light, sound, taste, smell and touch.
It is known that information about the stimulus is encoded in this pattern of action potentials and transmitted into and around the brain.
Although action potentials can vary somewhat in duration, amplitude and shape, they are typically treated as identical stereotyped events in neural coding studies. If the brief duration of an action potential (about 1ms) is ignored, an action potential sequence, or spike train, can be characterized simply by a series of all-or-none point events in time.* [3] The lengths of interspike intervals
(ISIs) between two successive spikes in a spike train often vary, apparently randomly.* [4] The study of neural coding involves measuring and characterizing how stimulus attributes, such as light or sound intensity, or motor actions, such as the direction of an arm movement, are represented by neuron action potentials or spikes. In order to describe and analyze neuronal ﬁring, statistical methods and methods of probability theory and stochastic point processes have been widely applied.
With the development of large-scale neural recording and decoding technologies, researchers have begun to crack the neural code and already provided the ﬁrst glimpse into the real-time neural code as memory is formed and

6.2 Encoding and decoding

6.3 Coding schemes
A sequence, or 'train', of spikes may contain information based on diﬀerent coding schemes. In motor neurons, for example, the strength at which an innervated muscle is ﬂexed depends solely on the 'ﬁring rate', the average number of spikes per unit time (a 'rate code'). At the other end, a complex 'temporal code' is based on the precise timing of single spikes. They may be locked to an external stimulus such as in the visual* [10] and auditory system or be generated intrinsically by the neural circuitry.* [11]
Whether neurons use rate coding or temporal coding is a topic of intense debate within the neuroscience community, even though there is no clear deﬁnition of what these terms mean. In one theory, termed “neuroelectrodynamics”, the following coding schemes are all considered to be epiphenomena, replaced instead by molecular changes reﬂecting the spatial distribution of electric ﬁelds within neurons as a result of the broad electromagnetic spectrum of action potentials, and manifested in information as spike directivity.* [12]* [13]* [14]* [15]* [16]

39

40

6.3.1

CHAPTER 6. NEURAL CODING

Rate coding

the stimulus. In practice, to get sensible averages, several spikes should occur within the time window. Typical valThe rate coding model of neuronal ﬁring communication ues are T = 100 ms or T = 500 ms, but the duration may states that as the intensity of a stimulus increases, the also be longer or shorter.* [19] frequency or rate of action potentials, or “spike ﬁring” The spike-count rate can be determined from a single
, increases. Rate coding is sometimes called frequency trial, but at the expense of losing all temporal resolucoding. tion about variations in neural response during the course
Rate coding is a traditional coding scheme, assuming that most, if not all, information about the stimulus is contained in the ﬁring rate of the neuron. Because the sequence of action potentials generated by a given stimulus varies from trial to trial, neuronal responses are typically treated statistically or probabilistically. They may be characterized by ﬁring rates, rather than as speciﬁc spike sequences. In most sensory systems, the ﬁring rate increases, generally non-linearly, with increasing stimulus intensity.* [17] Any information possibly encoded in the temporal structure of the spike train is ignored. Consequently, rate coding is ineﬃcient but highly robust with respect to the ISI 'noise'.* [4]

of the trial. Temporal averaging can work well in cases where the stimulus is constant or slowly varying and does not require a fast reaction of the organism —and this is the situation usually encountered in experimental protocols. Real-world input, however, is hardly stationary, but often changing on a fast time scale. For example, even when viewing a static image, humans perform saccades, rapid changes of the direction of gaze. The image projected onto the retinal photoreceptors changes therefore every few hundred milliseconds.* [19]

Rate coding was originally shown by ED Adrian and
Y Zotterman in 1926.* [18] In this simple experiment diﬀerent weights were hung from a muscle. As the weight of the stimulus increased, the number of spikes recorded from sensory nerves innervating the muscle also increased. From these original experiments, Adrian and
Zotterman concluded that action potentials were unitary events, and that the frequency of events, and not individual event magnitude, was the basis for most interneuronal communication.

number of spikes (averaged over trials) appearing during a short interval between times t and t+Δt, divided by the duration of the interval. It works for stationary as well as for time-dependent stimuli. To experimentally measure the time-dependent ﬁring rate, the experimenter records from a neuron while stimulating with some input sequence. The same stimulation sequence is repeated several times and the neuronal response is reported in a Peri-Stimulus-Time Histogram (PSTH). The time t is measured with respect to the start of the stimulation sequence. The Δt must be large enough (typically in the range of one or a few milliseconds) so there are suﬃcient number of spikes within the interval to obtain a reliable estimate of the average. The number of occurrences of spikes nK (t;t+Δt) summed over all repetitions of the experiment divided by the number K of repetitions is a measure of the typical activity of the neuron between time t and t+Δt. A further division by the interval length Δt yields time-dependent ﬁring rate r(t) of the neuron, which is equivalent to the spike density of PSTH.

Despite its shortcomings, the concept of a spike-count rate code is widely used not only in experiments, but also in models of neural networks. It has led to the idea that a
During rate coding, precisely calculating ﬁring rate is very neuron transforms information about a single input variimportant. In fact, the term “ﬁring rate”has a few dif- able (the stimulus strength) into a single continuous outferent deﬁnitions, which refer to diﬀerent averaging pro- put variable (the ﬁring rate). cedures, such as an average over time or an average over several repetitions of experiment.
Time-dependent ﬁring rate
In rate coding, learning is based on activity-dependent synaptic weight modiﬁcations.
The time-dependent ﬁring rate is deﬁned as the average

In the following decades, measurement of ﬁring rates became a standard tool for describing the properties of all types of sensory or cortical neurons, partly due to the relative ease of measuring rates experimentally. However, this approach neglects all the information possibly contained in the exact timing of the spikes. During recent years, more and more experimental evidence has suggested that a straightforward ﬁring rate concept based on temporal averaging may be too simplistic to describe brain activity.* [4]

Spike-count rate

For suﬃciently small Δt, r(t)Δt is the average number of spikes occurring between times t and t+Δt over multiple trials. If Δt is small, there will never be more than one spike within the interval between t and t+Δt on any given trial. This means that r(t)Δt is also the fraction of trials on which a spike occurred between those times. Equivalently, r(t)Δt is the probability that a spike occurs during this time interval.

The Spike-count rate, also referred to as temporal average, is obtained by counting the number of spikes that appear during a trial and dividing by the duration of trial.
The length T of the time window is set by experimenter and depends on the type of neuron recorded from and As an experimental procedure, the time-dependent ﬁring

6.3. CODING SCHEMES

41

rate measure is a useful method to evaluate neuronal activity, in particular in the case of time-dependent stimuli.
The obvious problem with this approach is that it can not be the coding scheme used by neurons in the brain. Neurons can not wait for the stimuli to repeatedly present in an exactly same manner before generating response.

domness, or precisely timed groups of spikes (temporal patterns) are candidates for temporal codes.* [25] As there is no absolute time reference in the nervous system, the information is carried either in terms of the relative timing of spikes in a population of neurons or with respect to an ongoing brain oscillation.* [2]* [4]

Nevertheless, the experimental time-dependent ﬁring rate measure can make sense, if there are large populations of independent neurons that receive the same stimulus.
Instead of recording from a population of N neurons in a single run, it is experimentally easier to record from a single neuron and average over N repeated runs. Thus, the time-dependent ﬁring rate coding relies on the implicit assumption that there are always populations of neurons.

The temporal structure of a spike train or ﬁring rate evoked by a stimulus is determined both by the dynamics of the stimulus and by the nature of the neural encoding process. Stimuli that change rapidly tend to generate precisely timed spikes and rapidly changing ﬁring rates no matter what neural coding strategy is being used. Temporal coding refers to temporal precision in the response that does not arise solely from the dynamics of the stimulus, but that nevertheless relates to properties of the stimulus.
The interplay between stimulus and encoding dynamics makes the identiﬁcation of a temporal code diﬃcult.

6.3.2

Temporal coding

When precise spike timing or high-frequency ﬁring-rate ﬂuctuations are found to carry information, the neural code is often identiﬁed as a temporal code.* [20] A number of studies have found that the temporal resolution of the neural code is on a millisecond time scale, indicating that precise spike timing is a signiﬁcant element in neural coding.* [2]* [21]
Neurons exhibit high-frequency ﬂuctuations of ﬁringrates which could be noise or could carry information.
Rate coding models suggest that these irregularities are noise, while temporal coding models suggest that they encode information. If the nervous system only used rate codes to convey information, a more consistent, regular ﬁring rate would have been evolutionarily advantageous, and neurons would have utilized this code over other less robust options.* [22] Temporal coding supplies an alternate explanation for the “noise,”suggesting that it actually encodes information and aﬀects neural processing. To model this idea, binary symbols can be used to mark the spikes: 1 for a spike, 0 for no spike. Temporal coding allows the sequence 000111000111 to mean something diﬀerent from 001100110011, even though the mean ﬁring rate is the same for both sequences, at
6 spikes/10 ms.* [23] Until recently, scientists had put the most emphasis on rate encoding as an explanation for post-synaptic potential patterns. However, functions of the brain are more temporally precise than the use of only rate encoding seems to allow. In other words, essential information could be lost due to the inability of the rate code to capture all the available information of the spike train. In addition, responses are diﬀerent enough between similar (but not identical) stimuli to suggest that the distinct patterns of spikes contain a higher volume of information than is possible to include in a rate code.* [24]

In temporal coding, learning can be explained by activitydependent synaptic delay modiﬁcations.* [26] The modiﬁcations can themselves depend not only on spike rates
(rate coding) but also on spike timing patterns (temporal coding), i.e., can be a special case of spike-timingdependent plasticity.
The issue of temporal coding is distinct and independent from the issue of independent-spike coding. If each spike is independent of all the other spikes in the train, the temporal character of the neural code is determined by the behavior of time-dependent ﬁring rate r(t). If r(t) varies slowly with time, the code is typically called a rate code, and if it varies rapidly, the code is called temporal.
Temporal coding in sensory systems

For very brief stimuli, a neuron's maximum ﬁring rate may not be fast enough to produce more than a single spike. Due to the density of information about the abbreviated stimulus contained in this single spike, it would seem that the timing of the spike itself would have to convey more information than simply the average frequency of action potentials over a given period of time.
This model is especially important for sound localization, which occurs within the brain on the order of milliseconds. The brain must obtain a large quantity of information based on a relatively short neural response. Additionally, if low ﬁring rates on the order of ten spikes per second must be distinguished from arbitrarily close rate coding for diﬀerent stimuli, then a neuron trying to discriminate these two stimuli may need to wait for a second or more to accumulate enough information. This is not consistent with numerous organisms which are able to discriminate between stimuli in the time frame of milTemporal codes employ those features of the spiking ac- liseconds, suggesting that a rate code is not the only model
*
tivity that cannot be described by the ﬁring rate. For at work. [23] example, time to ﬁrst spike after the stimulus onset, To account for the fast encoding of visual stimuli, it has characteristics based on the second and higher statistical been suggested that neurons of the retina encode visual inmoments of the ISI probability distribution, spike ran- formation in the latency time between stimulus onset and

42 ﬁrst action potential, also called latency to ﬁrst spike.* [27]
This type of temporal coding has been shown also in the auditory and somato-sensory system. The main drawback of such a coding scheme is its sensitivity to intrinsic neuronal ﬂuctuations.* [28] In the primary visual cortex of macaques, the timing of the ﬁrst spike relative to the start of the stimulus was found to provide more information than the interval between spikes. However, the interspike interval could be used to encode additional information, which is especially important when the spike rate reaches its limit, as in high-contrast situations. For this reason, temporal coding may play a part in coding deﬁned edges rather than gradual transitions.* [29]
The mammalian gustatory system is useful for studying temporal coding because of its fairly distinct stimuli and the easily discernible responses of the organism.* [30]
Temporally encoded information may help an organism discriminate between diﬀerent tastants of the same category (sweet, bitter, sour, salty, umami) that elicit very similar responses in terms of spike count. The temporal component of the pattern elicited by each tastant may be used to determine its identity (e.g., the diﬀerence between two bitter tastants, such as quinine and denatonium). In this way, both rate coding and temporal coding may be used in the gustatory system – rate for basic tastant type, temporal for more speciﬁc diﬀerentiation.* [31]
Research on mammalian gustatory system has shown that there is an abundance of information present in temporal patterns across populations of neurons, and this information is diﬀerent from that which is determined by rate coding schemes. Groups of neurons may synchronize in response to a stimulus. In studies dealing with the front cortical portion of the brain in primates, precise patterns with short time scales only a few milliseconds in length were found across small populations of neurons which correlated with certain information processing behaviors. However, little information could be determined from the patterns; one possible theory is they represented the higher-order processing taking place in the brain.* [24]
As with the visual system, in mitral/tufted cells in the olfactory bulb of mice, ﬁrst-spike latency relative to the start of a sniﬃng action seemed to encode much of the information about an odor. This strategy of using spike latency allows for rapid identiﬁcation of and reaction to an odorant. In addition, some mitral/tufted cells have speciﬁc ﬁring patterns for given odorants. This type of extra information could help in recognizing a certain odor, but is not completely necessary, as average spike count over the course of the animal's sniﬃng was also a good identiﬁer.* [32] Along the same lines, experiments done with the olfactory system of rabbits showed distinct patterns which correlated with diﬀerent subsets of odorants, and a similar result was obtained in experiments with the locust olfactory system.* [23]

CHAPTER 6. NEURAL CODING
Temporal coding applications
The speciﬁcity of temporal coding requires highly reﬁned technology to measure informative, reliable, experimental data. Advances made in optogenetics allow neurologists to control spikes in individual neurons, oﬀering electrical and spatial single-cell resolution. For example, blue light causes the light-gated ion channel channelrhodopsin to open, depolarizing the cell and producing a spike.
When blue light is not sensed by the cell, the channel closes, and the neuron ceases to spike. The pattern of the spikes matches the pattern of the blue light stimuli. By inserting channelrhodopsin gene sequences into mouse DNA, researchers can control spikes and therefore certain behaviors of the mouse (e.g., making the mouse turn left).* [33] Researchers, through optogenetics, have the tools to eﬀect diﬀerent temporal codes in a neuron while maintaining the same mean ﬁring rate, and thereby can test whether or not temporal coding occurs in speciﬁc neural circuits.* [34]
Optogenetic technology also has the potential to enable the correction of spike abnormalities at the root of several neurological and psychological disorders.* [34] If neurons do encode information in individual spike timing patterns, key signals could be missed by attempting to crack the code while looking only at mean ﬁring rates.* [23] Understanding any temporally encoded aspects of the neural code and replicating these sequences in neurons could allow for greater control and treatment of neurological disorders such as depression, schizophrenia, and Parkinson's disease. Regulation of spike intervals in single cells more precisely controls brain activity than the addition of pharmacological agents intravenously.* [33]
Phase-of-ﬁring code
Phase-of-ﬁring code is a neural coding scheme that combines the spike count code with a time reference based on oscillations. This type of code takes into account a time label for each spike according to a time reference based on phase of local ongoing oscillations at low* [35] or high frequencies.* [36] A feature of this code is that neurons adhere to a preferred order of spiking, resulting in ﬁring sequence.* [37]
It has been shown that neurons in some cortical sensory areas encode rich naturalistic stimuli in terms of their spike times relative to the phase of ongoing network ﬂuctuations, rather than only in terms of their spike count.* [35]* [38] Oscillations reﬂect local ﬁeld potential signals. It is often categorized as a temporal code although the time label used for spikes is coarse grained.
That is, four discrete values for phase are enough to represent all the information content in this kind of code with respect to the phase of oscillations in low frequencies. Phase-of-ﬁring code is loosely based on the phase precession phenomena observed in place cells of the hippocampus. 6.3. CODING SCHEMES
Phase code has been shown in visual cortex to involve also high-frequency oscillations.* [37] Within a cycle of gamma oscillation, each neuron has it own preferred relative ﬁring time. As a result, an entire population of neurons generates a ﬁring sequence that has a duration of up to about 15 ms.* [37]

6.3.3

Population coding

Population coding is a method to represent stimuli by using the joint activities of a number of neurons. In population coding, each neuron has a distribution of responses over some set of inputs, and the responses of many neurons may be combined to determine some value about the inputs. From the theoretical point of view, population coding is one of a few mathematically well-formulated problems in neuroscience. It grasps the essential features of neural coding and yet, is simple enough for theoretic analysis.* [39] Experimental studies have revealed that this coding paradigm is widely used in the sensor and motor areas of the brain. For example, in the visual area medial temporal (MT), neurons are tuned to the moving direction.* [40] In response to an object moving in a particular direction, many neurons in MT ﬁre, with a noise-corrupted and bell-shaped activity pattern across the population. The moving direction of the object is retrieved from the population activity, to be immune from the ﬂuctuation existing in a single neuron’ signal. In one s classic example in the primary motor cortex, Apostolos
Georgopoulos and colleagues trained monkeys to move a joystick towards a lit target.* [41]* [42] They found that a single neuron would ﬁre for multiple target directions.
However it would ﬁre fastest for one direction and more slowly depending on how close the target was to the neuron's 'preferred' direction.
Kenneth Johnson originally derived that if each neuron represents movement in its preferred direction, and the vector sum of all neurons is calculated (each neuron has a ﬁring rate and a preferred direction), the sum points in the direction of motion. In this manner, the population of neurons codes the signal for the motion. This particular population code is referred to as population vector coding. This particular study divided the ﬁeld of motor physiologists between Evarts' “upper motor neuron”group, which followed the hypothesis that motor cortex neurons contributed to control of single muscles, and the Georgopoulos group studying the representation of movement directions in cortex.
Population coding has a number of advantages, including reduction of uncertainty due to neuronal variability and the ability to represent a number of diﬀerent stimulus attributes simultaneously. Population coding is also much faster than rate coding and can reﬂect changes in the stimulus conditions nearly instantaneously.* [43] Individual neurons in such a population typically have diﬀerent

43 but overlapping selectivities, so that many neurons, but not necessarily all, respond to a given stimulus.
Typically an encoding function has a peak value such that activity of the neuron is greatest if the perceptual value is close to the peak value, and becomes reduced accordingly for values less close to the peak value.
It follows that the actual perceived value can be reconstructed from the overall pattern of activity in the set of neurons. The Johnson/Georgopoulos vector coding is an example of simple averaging. A more sophisticated mathematical technique for performing such a reconstruction is the method of maximum likelihood based on a multivariate distribution of the neuronal responses.
These models can assume independence, second order correlations ,* [44] or even more detailed dependencies such as higher order maximum entropy models* [45] or copulas.* [46]
Correlation coding
The correlation coding model of neuronal ﬁring claims that correlations between action potentials, or “spikes”
, within a spike train may carry additional information above and beyond the simple timing of the spikes. Early work suggested that correlation between spike trains can only reduce, and never increase, the total mutual information present in the two spike trains about a stimulus feature.* [47] However, this was later demonstrated to be incorrect. Correlation structure can increase information content if noise and signal correlations are of opposite sign.* [48] Correlations can also carry information not present in the average ﬁring rate of two pairs of neurons. A good example of this exists in the pentobarbital-anesthetized marmoset auditory cortex, in which a pure tone causes an increase in the number of correlated spikes, but not an increase in the mean ﬁring rate, of pairs of neurons.* [49]
Independent-spike coding
The independent-spike coding model of neuronal ﬁring claims that each individual action potential, or “spike”
, is independent of each other spike within the spike train.* [50]* [51]
Position coding
A typical population code involves neurons with a Gaussian tuning curve whose means vary linearly with the stimulus intensity, meaning that the neuron responds most strongly (in terms of spikes per second) to a stimulus near the mean. The actual intensity could be recovered as the stimulus level corresponding to the mean of the neuron with the greatest response. However, the noise inherent in neural responses means that a maximum likelihood estimation function is more accurate.

44

CHAPTER 6. NEURAL CODING distributed across neurons. A major result in neural coding from Olshausen et al.* [53] is that sparse coding of natural images produces wavelet-like oriented ﬁlters that resemble the receptive ﬁelds of simple cells in the visual cortex. The capacity of sparse codes may be increased by simultaneous use of temporal coding, as found in the locust olfactory system.* [54]

Plot of typical position coding

Given a potentially large set of input patterns, sparse coding algorithms (e.g. Sparse Autoencoder) attempt to automatically ﬁnd a small number of representative patterns which, when combined in the right proportions, reproduce the original input patterns. The sparse coding for the input then consists of those representative patterns.
For example, the very large set of English sentences can be encoded by a small number of symbols (i.e. letters, numbers, punctuation, and spaces) combined in a particular order for a particular sentence, and so a sparse coding for English would be those symbols.
Linear Generative Model
Most models of sparse coding are based on the linear generative model.* [55] In this model, the symbols are combined in a linear fashion to approximate the input.

Neural responses are noisy and unreliable.

This type of code is used to encode continuous variables such as joint position, eye position, color, or sound frequency. Any individual neuron is too noisy to faithfully encode the variable using rate coding, but an entire population ensures greater ﬁdelity and precision. For a population of unimodal tuning curves, i.e. with a single peak, the precision typically scales linearly with the number of neurons. Hence, for half the precision, half as many neurons are required. In contrast, when the tuning curves have multiple peaks, as in grid cells that represent space, the precision of the population can scale exponentially with the number of neurons. This greatly reduces the number of neurons required for the same precision.* [52]

More formally, given a k-dimensional set of real⃗ numbered input vectors ξ ∈ Rk , the goal of sparse coding is to determine n k-dimensional basis vectors
⃗
⃗ b1 , . . . , bn ∈ Rk along with a sparse n-dimensional vector of weights or coeﬃcients ⃗ ∈ Rn for each input vecs tor, so that a linear combination of the basis vectors with proportions given by the coeﬃcients results in a close ap⃗ ∑n b proximation to the input vector: ξ ≈ j=1 sj⃗ j .* [56]

The codings generated by algorithms implementing a linear generative model can be classiﬁed into codings with soft sparseness and those with hard sparseness.* [55]
These refer to the distribution of basis vector coeﬃcients for typical inputs. A coding with soft sparseness has a smooth Gaussian-like distribution, but peakier than
Gaussian, with many zero values, some small absolute values, fewer larger absolute values, and very few very large absolute values. Thus, many of the basis vectors are active. Hard sparseness, on the other hand, indicates
6.3.4 Sparse coding that there are many zero values, no or hardly any small absolute values, fewer larger absolute values, and very
The sparse code is when each item is encoded by the few very large absolute values, and thus few of the bastrong activation of a relatively small set of neurons. For sis vectors are active. This is appealing from a metabolic each item to be encoded, this is a diﬀerent subset of all perspective: less energy is used when fewer neurons are available neurons. ﬁring.* [55]
As a consequence, sparseness may be focused on temporal sparseness “a relatively small number of time periods
(
are active”) or on the sparseness in an activated population of neurons. In this latter case, this may be deﬁned in one time period as the number of activated neurons relative to the total number of neurons in the population. This seems to be a hallmark of neural computations since compared to traditional computers, information is massively

Another measure of coding is whether it is critically complete or overcomplete. If the number of basis vectors n is equal to the dimensionality k of the input set, the coding is said to be critically complete. In this case, smooth changes in the input vector result in abrupt changes in the coeﬃcients, and the coding is not able to gracefully handle small scalings, small translations, or noise in the inputs. If, however, the number of basis vectors is larger

6.5. REFERENCES than the dimensionality of the input set, the coding is overcomplete. Overcomplete codings smoothly interpolate between input vectors and are robust under input noise.* [57] The human primary visual cortex is estimated to be overcomplete by a factor of 500, so that, for example, a 14 x 14 patch of input (a 196-dimensional space) is coded by roughly 100,000 neurons.* [55]

6.4 See also
• Models of neural computation
• Neural correlate
• Cognitive map
• Neural decoding
• Deep learning
• Autoencoder
• Vector quantization
• Binding problem
• Artiﬁcial neural network
• Grandmother cell
• Feature integration theory
• pooling
• Sparse distributed memory

6.5 References
[1] Brown EN, Kass RE, Mitra PP (May 2004). “Multiple neural spike train data analysis: state-of-the-art and future challenges”. Nat. Neurosci. 7 (5): 456–61. doi:10.1038/nn1228. PMID 15114358.
[2] Thorpe, S.J. (1990). “Spike arrival times: A highly eﬃcient coding scheme for neural networks”(PDF). In Eckmiller, R.; Hartmann, G.; Hauske, G. Parallel processing in neural systems and computers (PDF). North-Holland. pp. 91–94. ISBN 978-0-444-88390-2.
[3] Gerstner, Wulfram; Kistler, Werner M. (2002). Spiking
Neuron Models: Single Neurons, Populations, Plasticity.
Cambridge University Press. ISBN 978-0-521-89079-3.
[4] Stein RB, Gossen ER, Jones KE (May 2005). “Neuronal variability: noise or part of the signal?". Nat. Rev.
Neurosci. 6 (5): 389–97. doi:10.1038/nrn1668. PMID
15861181.
[5] The Memory Code. http://www.scientificamerican.com/ article/the-memory-code/ 45

[6] Chen, G; Wang, LP; Tsien, JZ (2009).
“Neural population-level memory traces in the mouse hippocampus”. PLoS One.
4 (12): e8256. doi:10.1371/journal.pone.0008256. PMID 20016843.
[7] Zhang, H; Chen, G; Kuang, H; Tsien, JZ (Nov
2013). “Mapping and deciphering neural codes of
NMDA receptor-dependent fear memory engrams in the hippocampus”. PLoS One. 8 (11): e79454. doi:10.1371/journal.pone.0079454. PMID 24302990.
[8] Brain Decoding Project. org/ http://braindecodingproject.

[9] The Simons Collaboration on the Global Brain. http://www.simonsfoundation.org/life-sciences/ simons-collaboration-on-the-global-brain/
[10] Burcas G.T & Albright T.D. Gauging sensory representations in the brain. http://www.vcl.salk.edu/Publications/
PDF/Buracas_Albright_1999_TINS.pdf
[11] Gerstner W, Kreiter AK, Markram H, Herz AV; Kreiter;
Markram; Herz (November 1997). “Neural codes: ﬁring rates and beyond”. Proc. Natl. Acad. Sci. U.S.A.
94 (24): 12740–1. Bibcode:1997PNAS...9412740G. doi:10.1073/pnas.94.24.12740. PMC 34168. PMID
9398065.
[12] Aur D., Jog, MS., 2010 Neuroelectrodynamics: Understanding the brain language, IOS Press, 2010, doi:10.3233/978-1-60750-473-3-i [13] Aur, D.; Connolly, C.I.; Jog, M.S. (2005). “Computing spike directivity with tetrodes”. J. Neurosci 149 (1): 57–
63. doi:10.1016/j.jneumeth.2005.05.006.
[14] Aur, D.; Jog, M.S. (2007). “Reading the Neural Code:
What do Spikes Mean for Behavior?". Nature Precedings. doi:10.1038/npre.2007.61.1. [15] Fraser, A.; Frey, A. H. (1968). “Electromagnetic emission at micron wavelengths from active nerves”. Biophysical Journal 8 (6): 731–734. doi:10.1016/s00063495(68)86517-8.
[16] Aur, D (2012). “A comparative analysis of integrating visual information in local neuronal ensembles”
. Journal of neuroscience methods 207 (1): 23–30. doi:10.1016/j.jneumeth.2012.03.008. PMID 22480985.
[17] Kandel, E.; Schwartz, J.; Jessel, T.M. (1991). Principles of Neural Science (3rd ed.). Elsevier. ISBN 0444015620.
[18] Adrian ED & Zotterman Y. (1926). “The impulses produced by sensory nerve endings: Part II: The response of a single end organ.”. J Physiol (Lond.) 61: 151–171.
[19] http://icwww.epfl.ch/~{}gerstner/SPNM/node7.html
[20] Dayan, Peter; Abbott, L. F. (2001). Theoretical Neuroscience: Computational and Mathematical Modeling of
Neural Systems. Massachusetts Institute of Technology
Press. ISBN 978-0-262-04199-7.

46

CHAPTER 6. NEURAL CODING

[21] Butts DA, Weng C, Jin J et al.
(September
2007).
“Temporal precision in the neural code and the timescales of natural vision”.
Nature
449 (7158): 92–5.
Bibcode:2007Natur.449...92B.
doi:10.1038/nature06105. PMID 17805296.

[35] Montemurro MA, Rasch MJ, Murayama Y, Logothetis
NK, Panzeri S (March 2008). “Phase-of-ﬁring coding of natural visual stimuli in primary visual cortex”. Curr.
Biol. 18 (5): 375–80. doi:10.1016/j.cub.2008.02.023.
PMID 18328702.

[22] J. Leo van Hemmen, TJ Sejnowski. 23 Problems in Systems Neuroscience. Oxford Univ. Press, 2006. p.143158.

[36] Fries P, Nikolić D, Singer W (July 2007). “The gamma cycle”. Trends Neurosci. 30 (7): 309–16. doi:10.1016/j.tins.2007.05.005. PMID 17555828.

[23] Theunissen, F; Miller, JP (1995). “Temporal En- [37] Havenith MN, Yu S, Biederlack J, Chen NH, Singer coding in Nervous Systems: A Rigorous Deﬁnition”
W, Nikolić D (June 2011). “Synchrony makes neu. Journal of Computational Neuroscience 2: 149–162. rons ﬁre in sequence, and stimulus properties deterdoi:10.1007/bf00961885. mine who is ahead”. J. Neurosci. 31 (23): 8570–
84. doi:10.1523/JNEUROSCI.2817-10.2011. PMID
[24] Zador, Stevens, Charles, Anthony. “The enigma of the
21653861.
brain”. © Current Biology 1995, Vol 5 No 12. Retrieved
4/08/12. Check date values in: |accessdate= (help)
[38] Spike arrival times: A highly eﬃcient coding scheme for neural networks, SJ Thorpe - Parallel processing in neural
[25] Kostal L, Lansky P, Rospars JP (November 2007). systems, 1990
“Neuronal coding and spiking randomness”. Eur. J.
Neurosci. 26 (10): 2693–701. doi:10.1111/j.14609568.2007.05880.x. PMID 18001270.
[26] Geoﬀrois, E.; Edeline, J.M.; Vibert, J.F. (1994).“Learning by Delay Modiﬁcations” In Eeckman, Frank H. Com. putation in Neurons and Neural Systems. Springer. pp.
133–8. ISBN 978-0-7923-9465-5.
[27] Gollisch, T.; Meister, M. (22 February 2008). “Rapid
Neural Coding in the Retina with Relative Spike
Latencies”.
Science 319 (5866):
1108–1111.
doi:10.1126/science.1149639.
[28] Wainrib, Gilles; Michèle, Thieullen; Khashayar, Pakdaman (7 April 2010). “Intrinsic variability of latency to ﬁrst-spike”. Biological Cybernetics 103 (1): 43–56. doi:10.1007/s00422-010-0384-8. [39] Wu S, Amari S, Nakahara H (May 2002). “Population coding and decoding in a neural ﬁeld: a computational study”. Neural Comput 14 (5): 999–1026. doi:10.1162/089976602753633367. PMID 11972905.
[40] Maunsell JH, Van Essen DC (May 1983). “Functional properties of neurons in middle temporal visual area of the macaque monkey. I. Selectivity for stimulus direction, speed, and orientation”. J. Neurophysiol. 49 (5): 1127–
47. PMID 6864242.
[41] Intro to Sensory Motor Systems Ch. 38 page 766
[42] Science. 1986 Sep 26;233(4771):1416-9

[43] Hubel DH, Wiesel TN (October 1959).“Receptive ﬁelds of single neurones in the cat's striate cortex”. J. Phys[29] Victor, Johnathan D (2005). “Spike train metrics” iol. (Lond.) 148 (3): 574–91. PMC 1363130. PMID
. Current Opinion in Neurobiology 15 (5): 585–592.
14403679.
doi:10.1016/j.conb.2005.08.002.
[44] Schneidman, E, Berry, MJ, Segev, R, Bialek, W (2006),
[30] Hallock, Robert M.; Di Lorenzo, Patricia M. (2006).
Weak Pairwise Correlations Imply Strongly Correlated Net“Temporal coding in the gustatory system”. Neurowork States in a Neural Population 440, Nature 440, 1007science & Biobehavioral Reviews 30 (8): 1145–1160.
1012, doi:10.1038/nature04701 doi:10.1016/j.neubiorev.2006.07.005. [31] Carleton, Alan; Accolla, Riccardo; Simon, Sidney A.
(2010). “Coding in the mammalian gustatory system”. Trends in Neurosciences 33 (7): 326–334. doi:10.1016/j.tins.2010.04.002. [32] Wilson, Rachel I (2008).
“Neural and behavioral mechanisms of olfactory perception”.
Current Opinion in Neurobiology 18 (4):
408–412.
doi:10.1016/j.conb.2008.08.015.
[33] Karl Diesseroth, Lecture.“Personal Growth Series: Karl
Diesseroth on Cracking the Neural Code.”Google Tech
Talks. November 21, 2008. http://www.youtube.com/ watch?v=5SLdSbp6VjM [45] Amari, SL (2001), Information Geometry on Hierarchy of
Probability Distributions, IEEE Transactions on Information Theory 47, 1701-1711, CiteSeerX: 10.1.1.46.5226
[46] Onken, A, Grünewälder, S, Munk, MHJ, Obermayer,
K (2009), Analyzing Short-Term Noise Dependencies of
Spike-Counts in Macaque Prefrontal Cortex Using Copulas and the Flashlight Transformation, PLoS Comput Biol
5(11): e1000577, doi:10.1371/journal.pcbi.1000577
[47] Johnson, KO (Jun 1980). J Neurophysiol 43 (6): 1793–
815. Missing or empty |title= (help)
[48] Panzeri, Schultz, Treves, Rolls, Proc Biol Sci. 1999 May
22;266(1423):1001-12.

[34] Han X, Qian X, Stern P, Chuong AS, Boyden ES. “Informational lesions: optical perturbations of spike timing [49] Nature 381 (6583): doi:10.1038/381610a0. and neural synchrony via microbial opsin gene fusions.”
(help)
Cambridge, MA: MIT Media Lad, 2009.

610–3.
Jun 1996.
Missing or empty |title=

6.6. FURTHER READING

[50] Dayan P & Abbott LF. Theoretical Neuroscience: Computational and Mathematical Modeling of Neural Systems.
Cambridge, Massachusetts: The MIT Press; 2001. ISBN
0-262-04199-5
[51] Rieke F, Warland D, de Ruyter van Steveninck R, Bialek
W. Spikes: Exploring the Neural Code. Cambridge, Massachusetts: The MIT Press; 1999. ISBN 0-262-68108-0
[52] Mathis A, Herz AV, Stemmler MB; Herz; Stemmler (July 2012).
“Resolution of nested neuronal representations can be exponential in the number of neurons”. Phys. Rev. Lett.
109
Bibcode:2012PhRvL.109a8103M.
(1):
018103. doi:10.1103/PhysRevLett.109.018103. PMID
23031134.
[53] Olshausen, Bruno A. “Emergence of simple-cell receptive ﬁeld properties by learning a sparse code for natural images.”Nature 381.6583 (1996): 607-609.
[54] Gupta, N; Stopfer, M (6 October 2014).
“A
temporal channel for information in sparse sensory coding.”. Current Biology 24 (19): 2247–56. doi:10.1016/j.cub.2014.08.021. PMID 25264257.
[55] Rehn, Martin; Sommer, Friedrich T. (2007). “A network that uses few active neurones to code visual input predicts the diverse shapes of cortical receptive ﬁelds”
(PDF). Journal of Computational Neuroscience 22: 135–
146. doi:10.1007/s10827-006-0003-9.
[56] Lee, Honglak; Battle, Alexis; Raina, Rajat; Ng, Andrew
Y. (2006). “Eﬃcient sparse coding algorithms” (PDF).
Advances in Neural Information Processing Systems.
[57] Olshausen, Bruno A.; Field, David J. (1997). “Sparse
Coding with an Overcomplete Basis Set: A Strategy Employed by V1?" (PDF). Vision Research 37 (23): 3311–
3325. doi:10.1016/s0042-6989(97)00169-7.

6.6 Further reading
• Földiák P, Endres D, Sparse coding, Scholarpedia,
3(1):2984, 2008.
• Dayan P & Abbott LF. Theoretical Neuroscience:
Computational and Mathematical Modeling of Neural Systems. Cambridge, Massachusetts: The MIT
Press; 2001. ISBN 0-262-04199-5
• Rieke F, Warland D, de Ruyter van Steveninck R,
Bialek W. Spikes: Exploring the Neural Code. Cambridge, Massachusetts: The MIT Press; 1999. ISBN
0-262-68108-0
• Olshausen, B. A.; Field, D. J. “Emergence of simple-cell receptive ﬁeld properties by learning a sparse code for natural images” Nature 381 (6583):
.
607–9. doi:10.1038/381607a0.

47
• Tsien, JZ. et al. (2014). “On initial Brain Activity Mapping of episodic and semantic memory code in the hippocampus”.
Neurobiology of Learning and Memory 105: 200–210. doi:10.1016/j.nlm.2013.06.019. Chapter 7

Word embedding
Word embedding is the collective name for a set of language modeling and feature learning techniques in natural language processing where words from the vocabulary (and possibly phrases thereof) are mapped to vectors of real numbers in a low dimensional space, relative to the vocabulary size (“continuous space”).
There are several methods for generating this mapping.
They include neural networks,* [1] dimensionality reduction on the word co-occurrence matrix,* [2] and explicit representation in terms of the context in which words appear.* [3]
Word and phrase embeddings, when used as the underlying input representation, have been shown to boost the performance in NLP tasks such as syntactic parsing* [4] and sentiment analysis.* [5]

7.1 See also
• Brown clustering

7.2 References
[1] Mikolov, Tomas; Sutskever, Ilya; Chen, Kai; Corrado,
Greg; Dean, Jeﬀrey (2013). “Distributed Representations of Words and Phrases and their Compositionality”. arXiv:1310.4546 [cs.CL].
[2] Lebret, Rémi; Collobert, Ronan (2013).
“Word
Emdeddings through Hellinger PCA”. arXiv:1312.5542
[cs.CL].
[3] Levy, Omer; Goldberg, Yoav.“Linguistic Regularities in
Sparse and Explicit Word Representations” (PDF). Proceedings of the Eighteenth Conference on Computational
Natural Language Learning, Baltimore, Maryland, USA,
June. Association for Computational Linguistics. 2014.
[4] Socher, Richard; Bauer, John; Manning, Christopher; Ng,
Andrew. “Parsing with compositional vector grammars”
(PDF). Proceedings of the ACL conference. 2013.
[5] Socher, Richard; Perelygin, Alex; Wu, Jean; Chuang, Jason; Manning, Chris; Ng, Andrew; Potts, Chris. “Recursive Deep Models for Semantic Compositionality Over a

48

Sentiment Treebank” (PDF). Conference on Empirical
Methods in Natural Language Processing.

Chapter 8

Deep belief network
In machine learning, a deep belief network (DBN) is a generative graphical model, or alternatively a type of deep neural network, composed of multiple layers of latent variables (“hidden units”), with connections between the layers but not between units within each layer.* [1]

next. This also leads to a fast, layer-by-layer unsupervised training procedure, where contrastive divergence is applied to each sub-network in turn, starting from the“lowest” of layers (the lowest visible layer being a training pair set).
The observation, due to Geoﬀrey Hinton's student YeeWhye Teh,* [2] that DBNs can be trained greedily, one layer at a time, has been called a breakthrough in deep learning.* [4]* :6

Hidden layer 3

8.1 Training algorithm
The training algorithm for DBNs proceeds as follows. Let
X be a matrix of inputs, regarded as a set of feature vectors.* [2]

Hidden layer 2

1. Train a restricted Boltzmann machine on X to obtain its weight matrix, W. Use this as the weight matrix for between the lower two layers of the network.
2. Transform X by the RBM to produce new data X', either by sampling or by computing the mean activation of the hidden units.

Hidden layer 1

3. Repeat this procedure with X ← X' for the next pair of layers, until the top two layers of the network are reached. Visible layer (observed)

8.2 See also
• Bayesian network

Schematic overview of a deep belief net. Arrows represent directed connections in the graphical model that the net represents.

• Deep learning

When trained on a set of examples in an unsupervised way, a DBN can learn to probabilistically reconstruct its inputs. The layers then act as feature detectors on inputs.* [1] After this learning step, a DBN can be further trained in a supervised way to perform classiﬁcation.* [2]

8.3 References

DBNs can be viewed as a composition of simple, unsupervised networks such as restricted Boltzmann machines (RBMs)* [1] or autoencoders,* [3] where each subnetwork's hidden layer serves as the visible layer for the
49

[1] Hinton, G. (2009).“Deep belief networks” Scholarpedia
.
4 (5): 5947. doi:10.4249/scholarpedia.5947.
[2] Hinton, G. E.; Osindero, S.; Teh, Y. W. (2006).
“A Fast Learning Algorithm for Deep Belief Nets”
(PDF). Neural Computation 18 (7): 1527–1554. doi:10.1162/neco.2006.18.7.1527. PMID 16764513.

50

[3] Yoshua Bengio; Pascal Lamblin; Dan Popovici; Hugh
Larochelle (2007). Greedy Layer-Wise Training of Deep
Networks (PDF). NIPS.
[4] Bengio, Y. (2009).“Learning Deep Architectures for AI”
(PDF). Foundations and Trends in Machine Learning 2. doi:10.1561/2200000006. 8.4 External links
• “Deep Belief Networks” Deep Learning Tutorials.
.
• “Deep Belief Network Example”. Deeplearning4j
Tutorials.

CHAPTER 8. DEEP BELIEF NETWORK

Chapter 9

Convolutional neural network
For other uses, see CNN (disambiguation).

volutional neural networks use relatively little preprocessing. This means that the network is responsible
In machine learning, a convolutional neural network for learning the ﬁlters that in traditional algorithms were hand-engineered. The lack of a dependence on prior(or CNN) is a type of feed-forward artiﬁcial neural network where the individual neurons are tiled in such knowledge and the existence of diﬃcult to design handengineered features is a major advantage for CNNs. a way that they respond to overlapping regions in the visual ﬁeld.* [1] Convolutional networks were inspired by biological processes* [2] and are variations of multilayer perceptrons which are designed to use minimal amounts 9.2 History of preprocessing.* [3] They are widely used models for image and video recognition.
The design of convolutional neural networks follows the discovery of visual mechanisms in living organisms. In our brain, the visual cortex contains lots of cells. These cells are responsible for detecting light in small, overlap9.1 Overview ping sub-regions of the visual ﬁeld, called receptive ﬁelds.
These cells act as local ﬁlters over the input space. The
When used for image recognition, convolutional neural more complex cells have larger receptive ﬁelds. A connetworks (CNNs) consist of multiple layers of small neu- volution operator is created to perform the same function ron collections which look at small portions of the input by all of these cells. image, called receptive ﬁelds. The results of these collec- Convolutional neural networks were introduced in a 1980 tions are then tiled so that they overlap to obtain a better paper by Kunihiko Fukushima.* [7]* [9] In 1988 they were representation of the original image; this is repeated for separately developed, with explicit parallel and trainable every such layer. Because of this, they are able to toler- convolutions for temporal signals, by Toshiteru Homma, ate translation of the input image.* [4] Convolutional netLes Atlas, and Robert J. Marks II.* [10] Their design was works may include local or global pooling layers, which later improved in 1998 by Yann LeCun, Léon Bottou, combine the outputs of neuron clusters.* [5]* [6] They also
Yoshua Bengio, and Patrick Haﬀner,* [11] generalized consist of various combinations of convolutional layers in 2003 by Sven Behnke,* [12] and simpliﬁed by Patrice and fully connected layers, with pointwise nonlinearity
Simard, David Steinkraus, and John C. Platt in the same applied at the end of or after each layer.* [7] It is inspired year.* [13] The famous LeNet-5 network can classify digby biological process. To avoid the situation that there exits successfully, which is applied to recognize checking ist billions of parameters if all layers are fully connected, numbers. However, given more complex problems the the idea of using a convolution operation on small regions, breadth and depth of the network will continue to inhas been introduced. One major advantage of convolu- crease which would become limited by computing retional networks is the use of shared weight in convolu- sources. The approach used LeNet did not perform well tional layers, which means that the same ﬁlter (weights with more complex problems. bank) is used for each pixel in the layer; this both reduces
With the rise of eﬃcient GPU computing, it has become required memory size and improves performance.* [3] possible to train larger networks. In 2006 several pubSome Time delay neural networks also use a very similar lications described more eﬃcient ways to train convoluarchitecture to convolutional neural networks, especially tional neural networks with more layers.* [14]* [15]* [16] those for image recognition and/or classiﬁcation tasks, In 2011, they were reﬁned by Dan Ciresan et al. and since the “tiling”of the neuron outputs can easily be were implemented on a GPU with impressive perforcarried out in timed stages in a manner useful for analy- mance results.* [5] In 2012, Dan Ciresan et al. signiﬁsis of images.* [8] cantly improved upon the best performance in the literaCompared to other image classiﬁcation algorithms, con- ture for multiple image databases, including the MNIST
51

52

CHAPTER 9. CONVOLUTIONAL NEURAL NETWORK

database, the NORB database, the HWDB1.0 dataset Dropout “layer”
(Chinese characters), the CIFAR10 dataset (dataset of
60000 32x32 labeled RGB images),* [7] and the ImageNet dataset.* [17]

9.3 Details
9.3.1

Backpropagation

When doing propagation, the momentum and weight decay values are chosen to reduce oscillation during stochastic gradient descent. See Backpropagation for more. 9.3.2

Diﬀerent types of layers

Convolutional layer
Unlike a hand-coded convolution kernel (Sobel, Prewitt,
Roberts), in a convolutional neural net, the parameters of each convolution kernel are trained by the backpropagation algorithm. There are many convolution kernels in each layer, and each kernel is replicated over the entire image with the same parameters. The function of the convolution operators is to extract diﬀerent features of the input. The capacity of a neural net varies, depending on the number of layers. The ﬁrst convolution layers will obtain the low-level features, like edges, lines and corners. The more layers the network has, the higher-level features it will get.

Since a fully connected layer occupies most of the parameters, it is prone to overﬁtting. The dropout method
*
[19] is introduced to prevent overﬁtting. That paper deﬁnes (the simplest form of) dropout as: “The only diﬀerence”, from “Learning algorithms developed for
Restricted_Boltzmann_machine such as Contrastive Divergence”, “is that r”, the number derived (usually by
Sigmoid) from the incoming sum to the neural node from other nodes, “is ﬁrst sampled and only the hidden units that are retained are used for training.” and“dropout can be seen as multiplying by a Bernoulli_distribution random variable rb that takes the value 1/p with probability p and 0 otherwise.”In other words, that simplest form of dropout is to take the chance and see if it happens or not, to observe if the neural node spikes/ﬁres (instantly), instead of just remembering that chance happened.
Dropout also signiﬁcantly improves the speed of training.
This makes model combination practical, even for deep neural nets. Dropout is performed randomly. In the input layer, the probability of dropping a neuron is between
0.5 and 1, while in the hidden layers, a probability of 0.5 is used. The neurons that are dropped out, will not contribute to the forward pass and back propagation. This is equivalent to decreasing the number of neurons. This will create neural networks with diﬀerent architectures, but all of those networks will share the same weights.

In Neural_coding#Sparse_coding bio neurons generally some react and some dont, at any moment of experiReLU layer ence, as we are not perfect machines, not to say brains have the same chances and layer shapes used in simulated
ReLU is the abbreviation of Rectiﬁed Linear Units. This
Dropout, but we learn anyways through it. is a layer of neurons that use the non-saturating activation function f(x)=max(0,x). It increases the nonlinear prop- The biggest contribution of the dropout method is that, erties of the decision function and of the overall network although it eﬀectively generates 2^n neural nets, and as without aﬀecting the receptive ﬁelds of the convolution such, allows for model combination, at test time, only a single network needs to be tested. This is accomplished layer. by performing the test with the un-thinned network, while
Other functions are used to increase nonlinearity. For multiplying the output weights of each neuron with the example the saturating hyperbolic tangent f(x)=tanh(x), probability of that neuron being retained (i.e. not dropped f(x)=|tanh(x)|, and the sigmoid function f(x)=(1+e^(-x) out). )^(−1). Compared to tanh units, the advantage of ReLU
*
is that the neural network trains several times faster. [18] The“2^n neural nets”is accuracy but is pigeonholed by far less than 2^n bits having come in during the life of the neuralnet. We could the same way say Lambda_calculus
Pooling layer takes exponential time if it werent for using base2 in memory instead of counting in Unary_numeral_system.
In order to reduce variance, pooling layers compute the So if each of 2^n abstract neural nets pushes at least 1 max or average value of a particular feature over a region bit through those weights, they must have taken exponenof the image. This will ensure that the same result will tially many turns sequentially since the bandwidth is not be obtained, even when image features have small trans- that wide. P_versus_NP_problem millenium prize relations. This is an important operation for object classi- mains unclaimed, even though there are great optimizations. ﬁcation and detection.

9.5. FINE-TUNING

53

Loss layer

images that included faces at various angles and orientations and a further 20 million images without faces. They
It can use diﬀerent loss functions for diﬀerent tasks. Soft- used batches of 128 images over 50,000 iterations.* [23] max loss is used for predicting a single class of K mutually exclusive classes. Sigmoid cross-entropy loss is used for predicting K independent probability values in [0,1]. 9.4.2 Video analysis
Euclidean loss is used for regressing to real-valued labels
Video is more complex than images since it has an[-inf,inf] other temporal dimension. The common way is to fuse the features of diﬀerent convolutional neural networks, which are responsible for spatial and temporal
9.4 Applications stream.* [24]* [25]

9.4.1

Image recognition

Convolutional neural networks are often used in image recognition systems. They have achieved an error rate of
0.23 percent on the MNIST database, which as of February 2012 is the lowest achieved on the database.* [7] Another paper on using CNN for image classiﬁcation reported that the learning process was “surprisingly fast"; in the same paper, the best published results at the time were achieved in the MNIST database and the NORB database.* [5]
When applied to facial recognition, they were able to contribute to a large decrease in error rate.* [20] In another paper, they were able to achieve a 97.6 percent recognition rate on“5,600 still images of more than 10 subjects”
.* [2] CNNs have been used to assess video quality in an objective way after being manually trained; the resulting system had a very low root mean square error.* [8]
The ImageNet Large Scale Visual Recognition Challenge is a benchmark in object classiﬁcation and detection, with millions of images and hundreds of object classes. In the ILSVRC 2014, which is large-scale visual recognition challenge, almost every highly ranked team used CNN as their basic framework. The winner GoogLeNet* [21] increased the mean average precision of object detection to 0.439329, and reduced classiﬁcation error to 0.06656, the best result to date. Its network applied more than 30 layers. Performance of convolutional neural networks, on the ImageNet tests, is now close to that of humans.* [22]
The best algorithms still struggle with objects that are small or thin, such as a small ant on a stem of a ﬂower or a person holding a quill in their hand. They also have trouble with images that have been distorted with ﬁlters, an increasingly common phenomenon with modern digital cameras. By contrast, those kinds of images rarely trouble humans. Humans, however, tend to have trouble with other issues. For example, they are not good at classifying objects into ﬁne-grained categories such as the particular breed of dog or species of bird, whereas convolutional neural networks handle this with ease.
In 2015 a many-layered CNN demonstrated the ability to spot faces from a wide range of angles, including upside down, even when partially occluded with competitive performance. The network trained on a database of 200,000

9.4.3 Natural Language Processing
Convolutional neural networks have also seen use in the ﬁeld of natural language processing or NLP. Like the image classiﬁcation problem, some NLP tasks can be formulated as assigning labels to words in a sentence. The neural network trained raw material fashion will extract the features of the sentences. Using some classiﬁers, it could predict new sentences.* [26]

9.4.4 Playing Go
Convolutional neural networks have been used in computer Go. In December 2014, Christopher Clark and
Amos Storkey published a paper showing a convolutional network trained by supervised learning from a database of human professional games could outperform Gnu Go and win some games against Monte Carlo tree search Fuego
1.1 in a fraction of the time it took Fuego to play.* [27]
Shortly after it was announced that a large 12-layer convolutional neural network had correctly predicted the professional move in 55% of positions, equalling the accuracy of a 6 dan human player. When the trained convolutional network was used directly to play games of Go, without any search, it beat the traditional search program
GNU Go in 97% of games, and matched the performance of the Monte Carlo tree search program Fuego simulating ten thousand playouts (about a million positions) per move.* [28]

9.5 Fine-tuning
For many applications, only a small amount of training data is available. Convolutional neural networks usually require a large amount of training data in order to avoid over-ﬁtting. A common technique is to train the network on a larger data set from a related domain. Once the network parameters have converged an additional training step is performed using the in-domain data to ﬁne-tune the network weights. This allows convolutional networks to be successfully applied to problems with small training sets. 54

9.6 Common libraries
• Caﬀe: Caﬀe (replacement of Decaf) has been the most popular library for convolutional neural networks. It is created by the Berkeley Vision and
Learning Center (BVLC). The advantages are that it has cleaner architecture and faster speed. It supports both CPU and GPU, easily switching between them.
It is developed in C++, and has Python and MATLAB wrappers. In the developing of Caﬀe, protobuf is used to make researchers tune the parameters easily as well as adding or removing layers.
• Torch7 (www.torch.ch)
• OverFeat
• Cuda-convnet
• MatConvnet
• Theano: written in Python, using scientiﬁc python
• Deeplearning4j: Deep learning in Java and Scala on
GPU-enabled Spark

9.7 See also
• Neocognitron
• Convolution
• Deep learning
• Time delay neural network

9.8 References
[1] “Convolutional Neural Networks (LeNet) - DeepLearning 0.1 documentation”. DeepLearning 0.1. LISA Lab.
Retrieved 31 August 2013.
[2] Matusugu, Masakazu; Katsuhiko Mori; Yusuke Mitari;
Yuji Kaneda (2003). “Subject independent facial expression recognition with robust face detection using a convolutional neural network” (PDF). Neural Networks
16 (5): 555–559. doi:10.1016/S0893-6080(03)00115-1.
Retrieved 17 November 2013.

CHAPTER 9. CONVOLUTIONAL NEURAL NETWORK

[5] Ciresan, Dan; Ueli Meier; Jonathan Masci; Luca M. Gambardella; Jurgen Schmidhuber (2011). “Flexible, High
Performance Convolutional Neural Networks for Image
Classiﬁcation”(PDF). Proceedings of the Twenty-Second international joint conference on Artiﬁcial IntelligenceVolume Volume Two 2: 1237–1242. Retrieved 17
November 2013.
[6] Krizhevsky, Alex. “ImageNet Classiﬁcation with Deep
Convolutional Neural Networks” (PDF). Retrieved 17
November 2013.
[7] Ciresan, Dan; Meier, Ueli; Schmidhuber, Jürgen
(June 2012).
“Multi-column deep neural net2012 IEEE works for image classiﬁcation”.
Conference on Computer Vision and Pattern Recognition (New York, NY: Institute of Electrical and Electronics Engineers (IEEE)): 3642–3649. arXiv:1202.2745v1. doi:10.1109/CVPR.2012.6248110.
ISBN 9781467312264. OCLC 812295155. Retrieved
2013-12-09.
[8] Le Callet, Patrick; Christian Viard-Gaudin; Dominique
Barba (2006). “A Convolutional Neural Network
Approach for Objective Video Quality Assessment”
(PDF). IEEE Transactions on Neural Networks 17 (5):
1316–1327. doi:10.1109/TNN.2006.879766. PMID
17001990. Retrieved 17 November 2013.
[9] Fukushima, Kunihiko (1980). “Neocognitron: A Selforganizing Neural Network Model for a Mechanism of Pattern Recognition Unaﬀected by Shift in Position” (PDF). Biological Cybernetics 36 (4): 193–202. doi:10.1007/BF00344251. PMID 7370364. Retrieved
16 November 2013.
[10] Homma, Toshiteru; Les Atlas; Robert Marks II (1988).
“An Artiﬁcial Neural Network for Spatio-Temporal Bipolar Patters: Application to Phoneme Classiﬁcation”
(PDF). Advances in Neural Information Processing Systems 1: 31–40.
[11] LeCun, Yann; Léon Bottou; Yoshua Bengio; Patrick
Haﬀner (1998).“Gradient-based learning applied to document recognition” (PDF). Proceedings of the IEEE 86
(11): 2278–2324. doi:10.1109/5.726791. Retrieved 16
November 2013.
[12] S. Behnke. Hierarchical Neural Networks for Image Interpretation, volume 2766 of Lecture Notes in Computer
Science. Springer, 2003.
[13] Simard, Patrice, David Steinkraus, and John C. Platt.
“Best Practices for Convolutional Neural Networks Applied to Visual Document Analysis.”In ICDAR, vol. 3, pp. 958-962. 2003.

[3] LeCun, Yann.“LeNet-5, convolutional neural networks”
. Retrieved 16 November 2013.
[14] Hinton, GE; Osindero, S; Teh, YW (Jul 2006). “A fast learning algorithm for deep belief nets.” Neural computa.
[4] Korekado, Keisuke; Morie, Takashi; Nomura, Osamu; tion 18 (7): 1527–54. doi:10.1162/neco.2006.18.7.1527.
Ando, Hiroshi; Nakano, Teppei; Matsugu, Masakazu;
PMID 16764513.
Iwata, Atsushi (2003). “A Convolutional Neural Network VLSI for Image Recognition Using Merged/Mixed [15] Bengio, Yoshua; Lamblin, Pascal; Popovici, Dan;
Larochelle, Hugo (2007). “Greedy Layer-Wise TrainAnalog-Digital Architecture”. Knowledge-Based Inteling of Deep Networks”. Advances in Neural Information ligent Information and Engineering Systems: 169–176.
CiteSeerX: 10.1.1.125.3812.
Processing Systems: 153–160.

9.9. EXTERNAL LINKS

[16] Ranzato, MarcAurelio; Poultney, Christopher; Chopra,
Sumit; LeCun, Yann (2007). “Eﬃcient Learning of
Sparse Representations with an Energy-Based Model”
(PDF). Advances in Neural Information Processing Systems.
[17] 10. Deng, Jia, et al. “Imagenet: A large-scale hierarchical image database."Computer Vision and Pattern Recognition, 2009. CVPR 2009. IEEE Conference on. IEEE,
2009.
[18] Krizhevsky, A.; Sutskever, I.; Hinton, G. E. (2012).“Imagenet classiﬁcation with deep convolutional neural networks”. Advances in Neural Information Processing Systems 1: 1097–1105.
[19] Srivastava, Nitish; C. Geoﬀrey Hinton; Alex Krizhevsky;
Ilya Sutskever; Ruslan Salakhutdinov (2014). “Dropout:
A Simple Way to Prevent Neural Networks from overﬁtting” (PDF). Journal of Machine Learning Research 15
(1): 1929–1958.
[20] Lawrence, Steve; C. Lee Giles; Ah Chung Tsoi; Andrew
D. Back (1997). “Face Recognition: A Convolutional
Neural Network Approach”. Neural Networks, IEEE
Transactions on 8 (1): 98–113. doi:10.1109/72.554195.
CiteSeerX: 10.1.1.92.5813.
[21] Szegedy, Christian, et al. "Going deeper with convolutions.”arXiv preprint arXiv:1409.4842 (2014).
[22] O. Russakovsky et al., "ImageNet Large Scale Visual
Recognition Challenge", 2014.
[23] “The Face Detection Algorithm Set To Revolutionize Image Search”. Technology Review. February 16, 2015.
Retrieved February 2015.
[24] Karpathy, Andrej, et al. “Large-scale video classiﬁcation with convolutional neural networks.”
IEEE Conference on
Computer Vision and Pattern Recognition (CVPR). 2014.
[25] Simonyan, Karen, and Andrew Zisserman. “Two-stream convolutional networks for action recognition in videos.” arXiv preprint arXiv:1406.2199 (2014).
[26] Collobert, Ronan, and Jason Weston. “A uniﬁed architecture for natural language processing: Deep neural networks with multitask learning."Proceedings of the 25th international conference on Machine learning. ACM,
2008.
[27] Clark, C., Storkey A.J. (2014), "Teaching Deep Convolutional Networks to play Go".
[28] Maddison C.J., Huang A., Sutskever I., Silver D. (2014),
"Move evaluation in Go using deep convolutional neural networks". 9.9 External links
• A demonstration of a convolutional network created for character recognition
• Caﬀe

55
• Matlab toolbox
• MatConvnet
• Theano
• UFLDL Tutorial
• Deeplearning4j's Convolutional Nets

Chapter 10

Restricted Boltzmann machine class of Boltzmann machines, in particular the gradientbased contrastive divergence algorithm.* [7]

Hidden units
Visible units

Restricted Boltzmann machines can also be used in deep learning networks. In particular, deep belief networks can be formed by“stacking”RBMs and optionally ﬁnetuning the resulting deep network with gradient descent and backpropagation.* [8]

h3

v3 h4 v1

10.1 Structure h1 The standard type of RBM has binary-valued
(Boolean/Bernoulli) hidden and visible units, and consists of a matrix of weights W = (wi,j ) (size m×n) associated with the connection between hidden unit hj and visible unit vi , as well as bias weights (oﬀsets) ai for the visible units and bj for the hidden units. Given these, the energy of a conﬁguration (pair of boolean vectors)
(v,h) is deﬁned as

v2 h2 Diagram of a restricted Boltzmann machine with three visible units and four hidden units (no bias units).

A restricted Boltzmann machine (RBM) is a generative stochastic artiﬁcial neural network that can learn a probability distribution over its set of inputs.
RBMs were initially invented under the name Harmonium by Paul Smolensky in 1986,* [1] but only rose to prominence after Geoﬀrey Hinton and collaborators invented fast learning algorithms for them in the mid2000s. RBMs have found applications in dimensionality reduction,* [2] classiﬁcation,* [3] collaborative ﬁltering,* [4] feature learning* [5] and topic modelling.* [6]
They can be trained in either supervised or unsupervised ways, depending on the task.
As their name implies, RBMs are a variant of Boltzmann machines, with the restriction that their neurons must form a bipartite graph: a pair of nodes from each of the two groups of units, commonly referred to as the
“visible”
and “hidden”units respectively, may have a symmetric connection between them, and there are no connections between nodes within a group. By contrast, “unrestricted”Boltzmann machines may have connections between hidden units. This restriction allows for more eﬃcient training algorithms than are available for the general

E(v, h) = −

∑

a i vi −

i

∑

bj hj −

j

∑∑ i vi wi,j hj

j

or, in matrix notation,

E(v, h) = −aT v − bT h − v T W h
This energy function is analogous to that of a Hopﬁeld network. As in general Boltzmann machines, probability distributions over hidden and/or visible vectors are deﬁned in terms of the energy function:* [9]

P (v, h) =

1 −E(v,h) e Z

where Z is a partition function deﬁned as the sum of e−E(v,h) over all possible conﬁgurations (in other words, just a normalizing constant to ensure the probability distribution sums to 1). Similarly, the (marginal) probability of a visible (input) vector of booleans is the sum over all possible hidden layer conﬁgurations:* [9]

56

10.2. TRAINING ALGORITHM

P (v) =

1 ∑ −E(v,h) e Z

57
V (a matrix, each row of which is treated as a visible vector v ),

h

arg max

∏

P (v)
Since the RBM has the shape of a bipartite graph, with
W
v∈V no intra-layer connections, the hidden unit activations are mutually independent given the visible unit activations or equivalently, to maximize the expected log probability and conversely, the visible unit activations are mutually of V :* [10]* [11] independent given the hidden unit activations.* [7] That is, for m visible units and n hidden units, the conditional
[
] probability of a conﬁguration of the visible units v, given
∑
arg max E log P (v) a conﬁguration of the hidden units h, is
W

P (v|h) =

m
∏

P (vi |h)

i=1

Conversely, the conditional probability of h given v is

P (h|v) =

n
∏

P (hj |v)

j=1

The individual activation probabilities are given by
∑m
P (hj = 1|v) = ( (bj + i=1 wi,j vi ) and σ )
∑n
P (vi = 1|h) = σ ai + j=1 wi,j hj where σ denotes the logistic sigmoid.
The visible units of RBM can be multinomial, although the hidden units are Bernoulli. In this case, the logistic function for visible units is replaced by the Softmax function v∈V

The algorithm most often used to train RBMs, that is, to optimize the weight vector W , is the contrastive divergence (CD) algorithm due to Hinton, originally developed to train PoE (product of experts) models.* [13] * [14] The algorithm performs Gibbs sampling and is used inside a gradient descent procedure (similar to the way backpropagation is used inside such a procedure when training feedforward neural nets) to compute weight update.
The basic, single-step contrastive divergence (CD-1) procedure for a single sample can be summarized as follows:
1. Take a training sample v, compute the probabilities of the hidden units and sample a hidden activation vector h from this probability distribution.
2. Compute the outer product of v and h and call this the positive gradient.
3. From h, sample a reconstruction v' of the visible units, then resample the hidden activations h' from this. (Gibbs sampling step)
4. Compute the outer product of v' and h' and call this the negative gradient.

k
P (vi = 1|h) =

k exp(ak + Σj Wij hj ) i ΣK k=1 k exp(ak + Σj Wij hj ) i 5. Let the weight update to wi,j be the positive gradient minus the negative gradient, times some learning rate: ∆wi,j = ϵ(vhT − v ′ h′T ) .

where K is the number of discrete values that the visible values have. They are applied in Topic Modeling,* [6] and
The update rule for the biases a, b is deﬁned analogously.
RecSys.* [4]
A Practical Guide to Training RBMs written by Hinton can be found in his homepage.* [9]

10.1.1

Relation to other models

Restricted Boltzmann machines are a special case of Boltzmann machines and Markov random ﬁelds.* [10]* [11] Their graphical model corresponds to that of factor analysis.* [12]

10.2 Training algorithm
Restricted Boltzmann machines are trained to maximize the product of probabilities assigned to some training set

A restricted/layered Boltzmann machine (RBM) has either bit or scalar node values, an array for each layer, and between those are scalar values potentially for every pair of nodes one from each layer and an adjacent layer. It is run and trained using “weighted coin ﬂips”of a chance calculated at each individual node. Those chances are the logistic sigmoid of the sum of scalar weights of whichever pairs of nodes are on at the time, divided by temperature which decreases in each round of Simulated annealing as potentially all the data is trained in again. If either node in a pair is oﬀ, that weight is not counted. To run it, you go up and down the layers, updating the chances

58

CHAPTER 10. RESTRICTED BOLTZMANN MACHINE

and weighted coin ﬂips, until it converges to the coins in [9] Geoﬀrey Hinton (2010). A Practical Guide to Training Restricted Boltzmann Machines. UTML TR 2010–003, Unilowest layer (visible nodes) staying mostly a certain way. versity of Toronto.
To train it, it is the same shape as running it except you observe the weights of the pairs that are on, the ﬁrst time [10] Sutskever, Ilya; Tieleman, Tijmen (2010). “On the conup you add the learning rate between those pairs, then go vergence properties of contrastive divergence” (PDF). back down and up again and that time subtract the learnProc. 13th Int'l Conf. on AI and Statistics (AISTATS). ing rate. As Geoﬀrey Hinton explained it, the ﬁrst time up is to learn the data, and the second time up is to unlearn [11] Asja Fischer and Christian Igel. Training Restricted
Boltzmann Machines: An Introduction. Pattern Recogwhatever its earlier reaction was to the data. nition 47, pp. 25-39, 2014

10.3 See also
• Autoencoder
• Deep learning
• Helmholtz machine
• Hopﬁeld network

[12] María Angélica Cueto; Jason Morton; Bernd Sturmfels (2010).
“Geometry of the restricted Boltzmann machine” (PDF). Algebraic Methods in Statistics and Probability (American Mathematical Society) 516. arXiv:0908.4425. [13] Geoﬀrey Hinton (1999). Products of Experts. ICANN
1999.
[14] Hinton, G. E. (2002). “Training Products of Experts by Minimizing Contrastive Divergence”
(PDF). Neural Computation 14 (8):
1771–1800.
doi:10.1162/089976602760128018. PMID 12180402.

10.4 References
[1] Smolensky, Paul (1986). “Chapter 6: Information Processing in Dynamical Systems: Foundations of Harmony
Theory” (PDF). In Rumelhart, David E.; McLelland,
James L. Parallel Distributed Processing: Explorations in the Microstructure of Cognition, Volume 1: Foundations.
MIT Press. pp. 194–281. ISBN 0-262-68053-X.
[2] Hinton, G. E.; Salakhutdinov, R. R. (2006).
“Reducing the Dimensionality of Data with Neural
Networks” (PDF). Science 313 (5786): 504–507. doi:10.1126/science.1127647. PMID 16873662.
[3] Larochelle, H.; Bengio, Y. (2008).
Classiﬁcation
using discriminative restricted Boltzmann machines
(PDF). Proceedings of the 25th international conference on Machine learning - ICML '08. p. 536. doi:10.1145/1390156.1390224. ISBN 9781605582054.
[4] Salakhutdinov, R.; Mnih, A.; Hinton, G. (2007).
Restricted Boltzmann machines for collaborative ﬁltering. Proceedings of the 24th international conference on Machine learning - ICML '07. p. 791. doi:10.1145/1273496.1273596. ISBN 9781595937933.
[5] Coates, Adam; Lee, Honglak; Ng, Andrew Y. (2011). An analysis of single-layer networks in unsupervised feature learning (PDF). International Conference on Artiﬁcial Intelligence and Statistics (AISTATS).
[6] Ruslan Salakhutdinov and Geoﬀrey Hinton (2010).
Replicated softmax: an undirected topic model. Neural
Information Processing Systems 23.
[7] Miguel Á. Carreira-Perpiñán and Geoﬀrey Hinton (2005).
On contrastive divergence learning. Artiﬁcial Intelligence and Statistics.
[8] Hinton, G. (2009).“Deep belief networks” Scholarpedia
.
4 (5): 5947. doi:10.4249/scholarpedia.5947.

10.5 External links
• Introduction to Restricted Boltzmann Machines.
Edwin Chen's blog, July 18, 2011.
• Understanding RBMs. Deeplearning4j Documentation, December 29, 2014.

Chapter 11

Recurrent neural network
Not to be confused with Recursive neural network.

viding target signals for the RNN, instead a ﬁtness function or reward function is occasionally used to evaluate the RNN's performance, which is inﬂuencing its input stream through output units connected to actuators affecting the environment. Again, compare the section on training algorithms below.

A recurrent neural network (RNN) is a class of artiﬁcial neural network where connections between units form a directed cycle. This creates an internal state of the network which allows it to exhibit dynamic temporal behavior. Unlike feedforward neural networks, RNNs can use their internal memory to process arbitrary sequences
11.1.2 Hopﬁeld network of inputs. This makes them applicable to tasks such as unsegmented connected handwriting recognition, where The Hopﬁeld network is of historic interest although it is they have achieved the best known results.* [1] not a general RNN, as it is not designed to process sequences of patterns. Instead it requires stationary inputs.
It is a RNN in which all connections are symmetric. Invented by John Hopﬁeld in 1982, it guarantees that its dy11.1 Architectures namics will converge. If the connections are trained using
Hebbian learning then the Hopﬁeld network can perform
11.1.1 Fully recurrent network as robust content-addressable memory, resistant to connection alteration.
This is the basic architecture developed in the 1980s: a
A variation on the Hopﬁeld network is the bidirectional network of neuron-like units, each with a directed conassociative memory (BAM). The BAM has two layers, nection to every other unit. Each unit has a time-varying either of which can be driven as an input, to recall an real-valued activation. Each connection has a modiﬁable association and produce an output on the other layer.* [2] real-valued weight. Some of the nodes are called input nodes, some output nodes, the rest hidden nodes. Most architectures below are special cases.
11.1.3 Elman networks and Jordan netFor supervised learning in discrete time settings, training sequences of real-valued input vectors become sequences of activations of the input nodes, one input vector at a time. At any given time step, each non-input unit computes its current activation as a nonlinear function of the weighted sum of the activations of all units from which it receives connections. There may be teacher-given target activations for some of the output units at certain time steps. For example, if the input sequence is a speech signal corresponding to a spoken digit, the ﬁnal target output at the end of the sequence may be a label classifying the digit. For each sequence, its error is the sum of the deviations of all target signals from the corresponding activations computed by the network. For a training set of numerous sequences, the total error is the sum of the errors of all individual sequences. Algorithms for minimizing this error are mentioned in the section on training algorithms below.

works
The following special case of the basic architecture above was employed by Jeﬀ Elman. A three-layer network is used (arranged vertically as x, y, and z in the illustration), with the addition of a set of “context units”(u in the illustration). There are connections from the middle (hidden) layer to these context units ﬁxed with a weight of one.* [3] At each time step, the input is propagated in a standard feed-forward fashion, and then a learning rule is applied. The ﬁxed back connections result in the context units always maintaining a copy of the previous values of the hidden units (since they propagate over the connections before the learning rule is applied). Thus the network can maintain a sort of state, allowing it to perform such tasks as sequence-prediction that are beyond the power of a standard multilayer perceptron.

Jordan networks, due to Michael I. Jordan, are similar to
In reinforcement learning settings, there is no teacher pro- Elman networks. The context units are however fed from
59

60

CHAPTER 11. RECURRENT NEURAL NETWORK dict or label each element of the sequence based on both the past and the future context of the element. This is done by adding the outputs of two RNN, one processing the sequence from left to right, the other one from right to left. The combined outputs are the predictions of the teacher-given target signals. This technique proved to be especially useful when combined with LSTM RNN.* [10]

11.1.7 Continuous-time RNN
A continuous time recurrent neural network (CTRNN) is a dynamical systems model of biological neural networks.
A CTRNN uses a system of ordinary diﬀerential equations to model the eﬀects on a neuron of the incoming spike train. CTRNNs are more computationally eﬃcient than directly simulating every spike in a network as they do not model neural activations at this level of detail .

The Elman SRN

For a neuron i in the network with action potential yi the rate of change of activation is given by:

n
∑
the output layer instead of the hidden layer. The context τi yi = −yi + σ(
˙
wji yj − Θj ) + Ii (t) units in a Jordan network are also referred to as the state j=1 layer, and have a recurrent connection to themselves with no other nodes on this connection.* [3] Elman and Jordan
Where:
networks are also known as“simple recurrent networks”
(SRN).
• τi : Time constant of postsynaptic node

11.1.4

Echo state network

The echo state network (ESN) is a recurrent neural network with a sparsely connected random hidden layer. The weights of output neurons are the only part of the network that can change and be trained. ESN are good at reproducing certain time series.* [4] A variant for spiking neurons is known as Liquid state machines.* [5]

• yi : Activation of postsynaptic node
• yi : Rate of change of activation of postsynaptic
˙
node
• wji : Weight of connection from pre to postsynaptic node • σ(x) : Sigmoid of x e.g. σ(x) = 1/(1 + e−x ) .

11.1.5

Long short term memory network

The Long short term memory (LSTM) network, developed by Hochreiter & Schmidhuber in 1997,* [6] is an artiﬁcial neural net structure that unlike traditional RNNs doesn't have the vanishing gradient problem (compare the section on training algorithms below). It works even when there are long delays, and it can handle signals that have a mix of low and high frequency components.
LSTM RNN outperformed other methods in numerous applications such as language learning* [7] and connected handwriting recognition.* [8]

• yj : Activation of presynaptic node
• Θj : Bias of presynaptic node
• Ii (t) : Input (if any) to node
CTRNNs have frequently been applied in the ﬁeld of evolutionary robotics, where they have been used to address, for example, vision,* [11] co-operation* [12] and minimally cognitive behaviour.* [13]

11.1.8 Hierarchical RNN
11.1.6

Bi-directional RNN

There are many instances of hierarchical RNN whose elInvented by Schuster & Paliwal in 1997,* [9] bi- ements are connected in various ways to decompose hidirectional RNN or BRNN use a ﬁnite sequence to pre- erarchical behavior into useful subprograms.* [14]* [15]

11.2. TRAINING

11.1.9

Recurrent multilayer perceptron

Generally, a Recurrent Multi-Layer Perceptron (RMLP) consists of a series of cascaded subnetworks, each of which consists of multiple layers of nodes. Each of these subnetworks is entirely feed-forward except for the last layer, which can have feedback connections among itself.
Each of these subnets is connected only by feed forward connections.* [16]

11.1.10

61

11.1.14 Bidirectional Associative Memory
(BAM)
First introduced by Kosko,* [21] BAM neural networks store associative data as a vector. The bi-directionality comes from passing information through a matrix and its transpose. Typically, bipolar encoding is preferred to binary encoding of the associative pairs. Recently, stochastic BAM models using Markov stepping were optimized for increased network stability and relevance to real-world applications.* [22]

Second Order Recurrent Neural
Network

11.2 Training
Second order RNNs use higher order weights wijk instead of the standard wij weights, and inputs and states can be a product. This allows a direct mapping to a 11.2.1 Gradient descent ﬁnite state machine both in training and in representation* [17]* [18] Long short term memory is an example of To minimize total error, gradient descent can be used to change each weight in proportion to the derivative of the this. error with respect to that weight, provided the non-linear activation functions are diﬀerentiable. Various methods for doing so were developed in the 1980s and early 1990s by Paul Werbos, Ronald J. Williams, Tony Robinson,
11.1.11 Multiple Timescales Recurrent Jürgen Schmidhuber, Sepp Hochreiter, Barak PearlmutNeural Network (MTRNN) Model ter, and others.
MTRNN considered as a possible neural-based computational model that imitates to some extent, the brain activities.* [19] It has ability to simulate the functional hierarchy of the brain through self-organization that is not only depended on the spatial connection among neurons, but also on distinct types of neuron activities, each with distinct time properties. With such varied neuron activities, continuous sequences of any set of behavior are segmented into reusable primitives, which in turn are ﬂexibly integrated into diverse sequential behaviors. The biological approval of such a type of hierarchy has been discussed on the memory-prediction theory of brain function by Jeﬀ Hawkins in his book On Intelligence.

11.1.12

11.1.13

The standard method is called "backpropagation through time" or BPTT, and is a generalization of backpropagation for feed-forward networks,* [23]* [24] and like that method, is an instance of Automatic diﬀerentiation in the reverse accumulation mode or Pontryagin's minimum principle. A more computationally expensive online variant is called "Real-Time Recurrent Learning" or RTRL,* [25]* [26] which is an instance of Automatic diﬀerentiation in the forward accumulation mode with stacked tangent vectors. Unlike BPTT this algorithm is local in time but not local in space.* [27]* [28]

There also is an online hybrid between BPTT and RTRL with intermediate complexity,* [29]* [30] and there are variants for continuous time.* [31] A major problem with gradient descent for standard RNN architectures is that error gradients vanish exponentially quickly with the size of the time lag between important events.* [32] * [33] The
Pollack’s sequential cascaded net- Long short term memory architecture together with a works BPTT/RTRL hybrid learning method was introduced in an attempt to overcome these problems.* [6]

Neural Turing Machines

NTMs are method of extending the capabilities of recurrent neural networks by coupling them to external memory resources, which they can interact with by attentional processes. The combined system is analogous to a Turing Machine or Von Neumann architecture but is diﬀerentiable end-to-end, allowing it to be eﬃciently trained with gradient descent.* [20]

11.2.2 Hessian Free Optimisation
Successful training on complex tasks has been achieved by employing Hessian Free Optimisation. The speedup compared with previous training methods now makes
RNN applications feasible.* [34]

62

11.2.3

CHAPTER 11. RECURRENT NEURAL NETWORK

Global optimization methods

Training the weights in a neural network can be modeled as a non-linear global optimization problem. A target function can be formed to evaluate the ﬁtness or error of a particular weight vector as follows: First, the weights in the network are set according to the weight vector.
Next, the network is evaluated against the training sequence. Typically, the sum-squared-diﬀerence between the predictions and the target values speciﬁed in the training sequence is used to represent the error of the current weight vector. Arbitrary global optimization techniques may then be used to minimize this target function.

tation for the current time step.
In particular, recurrent neural networks can appear as nonlinear versions of ﬁnite impulse response and inﬁnite impulse response ﬁlters and also as a nonlinear autoregressive exogenous model (NARX)* [38]

11.4 Issues with recurrent neural networks Most RNNs have had scaling issues. In particular, RNNs cannot be easily trained for large numbers of neuron units
The most common global optimization method for train- nor for large numbers of inputs units . Successful training ing RNNs is genetic algorithms, especially in unstruc- has been mostly in time series problems with few inputs tured networks.* [35]* [36]* [37] and in chemical process control.
Initially, the genetic algorithm is encoded with the neural network weights in a predeﬁned manner where one gene in the chromosome represents one weight link, hence- 11.5 References forth; the whole network is represented as a single chromosome. The ﬁtness function is evaluated as follows: 1)
[1] A. Graves, M. Liwicki, S. Fernandez, R. Bertolami, H. each weight encoded in the chromosome is assigned to the
Bunke, J. Schmidhuber. A Novel Connectionist System respective weight link of the network ; 2) the training set for Improved Unconstrained Handwriting Recognition. of examples is then presented to the network which propIEEE Transactions on Pattern Analysis and Machine Inagates the input signals forward ; 3) the mean-squaredtelligence, vol. 31, no. 5, 2009. error is returned to the ﬁtness function ; 4) this function
[2] Rául Rojas (1996). Neural networks: a systematic introwill then drive the genetic selection process.
There are many chromosomes that make up the population; therefore, many diﬀerent neural networks are evolved until a stopping criterion is satisﬁed. A common stopping scheme is: 1) when the neural network has learnt a certain percentage of the training data or 2) when the minimum value of the mean-squared-error is satisﬁed or 3) when the maximum number of training generations has been reached. The stopping criterion is evaluated by the ﬁtness function as it gets the reciprocal of the mean-squared-error from each neural network during training. Therefore, the goal of the genetic algorithm is to maximize the ﬁtness function, hence, reduce the meansquared-error.
Other global (and/or evolutionary) optimization techniques may be used to seek a good set of weights such as Simulated annealing or Particle swarm optimization.

11.3 Related ﬁelds and models
RNNs may behave chaotically. In such cases, dynamical systems theory may be used for analysis.
Recurrent neural networks are in fact recursive neural networks with a particular structure: that of a linear chain.
Whereas recursive neural networks operate on any hierarchical structure, combining child representations into parent representations, recurrent neural networks operate on the linear progression of time, combining the previous time step and a hidden representation into the represen-

duction. Springer. p. 336. ISBN 978-3-540-60505-8.
[3] Cruse, Holk; Neural Networks as Cybernetic Systems, 2nd and revised edition
[4] H. Jaeger. Harnessing nonlinearity: Predicting chaotic systems and saving energy in wireless communication.
Science, 304:78–80, 2004.
[5] W. Maass, T. Natschläger, and H. Markram. A fresh look at real-time computation in generic recurrent neural circuits. Technical report, Institute for Theoretical Computer Science, TU Graz, 2002.
[6] Hochreiter, Sepp; and Schmidhuber, Jürgen; Long ShortTerm Memory, Neural Computation, 9(8):1735–1780,
1997
[7] Gers, Felix A.; and Schmidhuber, Jürgen; LSTM Recurrent Networks Learn Simple Context Free and Context Sensitive Languages, IEEE Transactions on Neural Networks,
12(6):1333–1340, 2001
[8] Graves, Alex; and Schmidhuber, Jürgen; Oﬄine Handwriting Recognition with Multidimensional Recurrent Neural Networks, in Bengio, Yoshua; Schuurmans, Dale; Lafferty, John; Williams, Chris K. I.; and Culotta, Aron
(eds.), Advances in Neural Information Processing Systems
22 (NIPS'22), December 7th–10th, 2009, Vancouver, BC,
Neural Information Processing Systems (NIPS) Foundation, 2009, pp. 545–552
[9] Bidirectional recurrent neural networks. IEEE Transactions on Signal Processing, 45:2673–81, November 1997.

11.5. REFERENCES

[10] A. Graves and J. Schmidhuber. Framewise phoneme classiﬁcation with bidirectional LSTM and other neural network architectures. Neural Networks, 18:602–610, 2005.
[11] Harvey, Inman; Husbands, P. and Cliﬀ, D. (1994).“Seeing the light: Artiﬁcial evolution, real vision”. Proceedings of the third international conference on Simulation of adaptive behavior: from animals to animats 3: 392–401.
[12] Quinn, Matthew (2001).“Evolving communication without dedicated communication channels”. Advances in
Artiﬁcial Life. Lecture Notes in Computer Science 2159:
357–366. doi:10.1007/3-540-44811-X_38. ISBN 9783-540-42567-0.
[13] Beer, R.D. (1997). “The dynamics of adaptive behavior:
A research program”. Robotics and Autonomous Systems
20 (2–4): 257–289. doi:10.1016/S0921-8890(96)000632.
[14] J. Schmidhuber. Learning complex, extended sequences using the principle of history compression. Neural Computation, 4(2):234-242, 1992
[15] R.W. Paine, J. Tani, “How hierarchical control selforganizes in artiﬁcial adaptive systems,”Adaptive Behavior, 13(3), 211-225, 2005.
[16] “CiteSeerX —Recurrent Multilayer Perceptrons for Identiﬁcation and Control: The Road to Applications”. Citeseerx.ist.psu.edu. Retrieved 2014-01-03.

63

[25] A. J. Robinson and F. Fallside. The utility driven dynamic error propagation network. Technical Report CUED/FINFENG/TR.1, Cambridge University Engineering Department, 1987.
[26] R. J. Williams and D. Zipser. Gradient-based learning algorithms for recurrent networks and their computational complexity. In Back-propagation: Theory, Architectures and Applications. Hillsdale, NJ: Erlbaum, 1994.
[27] J. Schmidhuber. A local learning algorithm for dynamic feedforward and recurrent networks. Connection Science,
1(4):403–412, 1989.
[28] Neural and Adaptive Systems: Fundamentals through
Simulation. J.C. Principe, N.R. Euliano, W.C. Lefebvre
[29] J. Schmidhuber. A ﬁxed size storage O(n3) time complexity learning algorithm for fully recurrent continually running networks. Neural Computation, 4(2):243–248,
1992.
[30] R. J. Williams. Complexity of exact gradient computation algorithms for recurrent neural networks. Technical Report Technical Report NU-CCS-89-27, Boston: Northeastern University, College of Computer Science, 1989.
[31] B. A. Pearlmutter. Learning state space trajectories in recurrent neural networks. Neural Computation, 1(2):263–
269, 1989.
[32] S. Hochreiter. Untersuchungen zu dynamischen neuronalen Netzen. Diploma thesis, Institut f. Informatik,
Technische Univ. Munich, 1991.

[17] C.L. Giles, C.B. Miller, D. Chen, H.H. Chen, G.Z. Sun,
Y.C. Lee, “Learning and Extracting Finite State Automata with Second-Order Recurrent Neural Networks,” [33] S. Hochreiter, Y. Bengio, P. Frasconi, and J. SchmidNeural Computation, 4(3), p. 393, 1992. huber. Gradient ﬂow in recurrent nets: the diﬃculty of learning long-term dependencies. In S. C. Kremer and J.
[18] C.W. Omlin, C.L. Giles, “Constructing Deterministic
F. Kolen, editors, A Field Guide to Dynamical Recurrent
Finite-State Automata in Recurrent Neural Networks,”
Neural Networks. IEEE Press, 2001.
Journal of the ACM, 45(6), 937-972, 1996.
[34] Martens, James, and Ilya Sutskever. "Training deep and
[19] Y. Yamashita, J. Tani, “Emergence of functional recurrent networks with hessian-free optimization.”In hierarchy in a multiple timescale neural network
Neural Networks: Tricks of the Trade, pp. 479-535. model: a humanoid robot experiment,”PLoS ComSpringer Berlin Heidelberg, 2012. putational Biology, 4(11), e1000220, 211-225, 2008. http://journals.plos.org/ploscompbiol/article?id=10. [35] F. J. Gomez and R. Miikkulainen.
Solving non1371/journal.pcbi.1000220
Markovian control tasks with neuroevolution. Proc. IJCAI 99, Denver, CO, 1999. Morgan Kaufmann.
[20] http://arxiv.org/pdf/1410.5401v2.pdf
[36] Applying Genetic Algorithms to Recurrent Neural Net[21] Kosko, B. (1988). “Bidirectional associative memories” works for Learning Network Parameters and Architec. IEEE Transactions on Systems, Man, and Cybernetics 18 ture. O. Syed, Y. Takefuji
(1): 49–60. doi:10.1109/21.87054.
[37] F. Gomez, J. Schmidhuber, R. Miikkulainen. Ac[22] Rakkiyappan, R.; Chandrasekar, A.; Lakshmanan, S.; celerated Neural Evolution through Cooperatively CoePark, Ju H. (2 January 2015). “Exponential stabilvolved Synapses. Journal of Machine Learning Research ity for markovian jumping stochastic BAM neural net(JMLR), 9:937-965, 2008. works with mode-dependent probabilistic time-varying delays and impulse control”. Complexity 20 (3): 39–65. [38] Hava T. Siegelmann, Bill G. Horne, C. Lee Giles,“Computational capabilities of recurrent NARX neural netdoi:10.1002/cplx.21503. works,” IEEE Transactions on Systems, Man, and Cyber[23] P. J. Werbos. Generalization of backpropagation with apnetics, Part B 27(2): 208-215 (1997). plication to a recurrent gas market model. Neural Networks, 1, 1988.
[24] David E. Rumelhart; Geoﬀrey E. Hinton; Ronald J.
Williams. Learning Internal Representations by Error
Propagation.

• Mandic, D. & Chambers, J. (2001). Recurrent Neural Networks for Prediction: Learning Algorithms,
Architectures and Stability. Wiley. ISBN 0-47149517-4.

64
• Elman, J.L. (1990).
“Finding Structure in
Time”. Cognitive Science 14 (2): 179–211. doi:10.1016/0364-0213(90)90002-E. 11.6 External links
• RNNSharp CRFs based on recurrent neural networks (C#, .NET)
• Recurrent Neural Networks with over 60 RNN papers by Jürgen Schmidhuber's group at IDSIA
• Elman Neural Network implementation for WEKA

CHAPTER 11. RECURRENT NEURAL NETWORK

Chapter 12

Long short term memory
Long short-term memory (LSTM) is a recurrent neural network (RNN) architecture (an artiﬁcial neural network) published* [1] in 1997 by Sepp Hochreiter and Jürgen
Schmidhuber. Like most RNNs, an LSTM network is universal in the sense that given enough network units it can compute anything a conventional computer can compute, provided it has the proper weight matrix, which may be viewed as its program. Unlike traditional RNNs, an
LSTM network is well-suited to learn from experience to classify, process and predict time series when there are very long time lags of unknown size between important events. This is one of the main reasons why LSTM outperforms alternative RNNs and Hidden Markov Models and other sequence learning methods in numerous applications. For example, LSTM achieved the best known results in unsegmented connected handwriting recognition,* [2] and in 2009 won the ICDAR handwriting competition. LSTM networks have also been used for automatic speech recognition, and were a major component of a network that recently achieved a record 17.7% phoneme error rate on the classic TIMIT natural speech dataset.* [3]
A typical implementation of an LSTM block.

Π

Σ

Π

Π

12.1 Architecture
An LSTM network is an artiﬁcial neural network that contains LSTM blocks instead of, or in addition to, regular network units. An LSTM block may be described as a“smart” network unit that can remember a value for an arbitrary length of time. An LSTM block contains gates that determine when the input is signiﬁcant enough to remember, when it should continue to remember or forget the value, and when it should output the value.
A typical implementation of an LSTM block is shown to the right. The four units shown at the bottom of the ﬁg∑ ure are sigmoid units ( y = s( wi xi ) , where s is some squashing function, such as the logistic function). The left-most of these units computes a value which is conditionally fed as an input value to the block's memory. The other three units serve as gates to determine when values are allowed to ﬂow into or out of the block's memory. The second unit from the left (on the bottom row) is the “input gate”. When it outputs a value close to zero, it zeros

out the value from the left-most unit, eﬀectively blocking that value from entering into the next layer. The second unit from the right is the “forget gate”. When it outputs a value close to zero, the block will eﬀectively forget whatever value it was remembering. The right-most unit
(on the bottom row) is the“output gate”. It determines when the unit should output the value in its memory. The units containing the Π symbol compute the product of their inputs ( y = Πxi ). These units have no weights.
The unit with the Σ symbol computes a linear function
∑
of its inputs ( y = wi xi .) The output of this unit is not squashed so that it can remember the same value for many time-steps without the value decaying. This value is fed back in so that the block can “remember”it (as long as the forget gate allows). Typically, this value is also fed into the 3 gating units to help them make gating decisions. 65

66

12.2 Training
To minimize LSTM's total error on a set of training sequences, iterative gradient descent such as backpropagation through time can be used to change each weight in proportion to its derivative with respect to the error. A major problem with gradient descent for standard RNNs is that error gradients vanish exponentially quickly with the size of the time lag between important events, as ﬁrst realized in 1991.* [4]* [5] With LSTM blocks, however, when error values are back-propagated from the output, the error becomes trapped in the memory portion of the block. This is referred to as an “error carousel”, which continuously feeds error back to each of the gates until they become trained to cut oﬀ the value.
Thus, regular backpropagation is eﬀective at training an
LSTM block to remember values for very long durations.
LSTM can also be trained by a combination of artiﬁcial evolution for weights to the hidden units, and pseudoinverse or support vector machines for weights to the output units.* [6] In reinforcement learning applications
LSTM can be trained by policy gradient methods or evolution strategies or genetic algorithms.

12.3 Applications
Applications of LSTM include:
• Robot control* [7]
• Time series prediction* [8]
• Speech recognition* [9]* [10]* [11]
• Rhythm learning* [12]
• Music composition* [13]
• Grammar learning* [14]* [15]* [16]
• Handwriting recognition* [17]* [18]
• Human action recognition* [19]
• Protein Homology Detection* [20]

12.4 See also
• Artiﬁcial neural network
• Prefrontal Cortex Basal Ganglia Working Memory
(PBWM)
• Recurrent neural network
• Time series
• Long-term potentiation

CHAPTER 12. LONG SHORT TERM MEMORY

12.5 References
[1] S. Hochreiter and J. Schmidhuber. Long short-term memory. Neural Computation, 9(8):1735–1780, 1997.
[2] A. Graves, M. Liwicki, S. Fernandez, R. Bertolami, H.
Bunke, J. Schmidhuber. A Novel Connectionist System for Improved Unconstrained Handwriting Recognition.
IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 31, no. 5, 2009.
[3] Graves, Alex; Mohamed, Abdel-rahman; Hinton, Geoffrey (2013). “Speech Recognition with Deep Recurrent
Neural Networks”. Acoustics, Speech and Signal Processing (ICASSP), 2013 IEEE International Conference on:
6645–6649.
[4] S. Hochreiter. Untersuchungen zu dynamischen neuronalen Netzen. Diploma thesis, Institut f. Informatik,
Technische Univ. Munich, 1991.
[5] S. Hochreiter, Y. Bengio, P. Frasconi, and J. Schmidhuber. Gradient ﬂow in recurrent nets: the diﬃculty of learning long-term dependencies. In S. C. Kremer and J.
F. Kolen, editors, A Field Guide to Dynamical Recurrent
Neural Networks. IEEE Press, 2001.
[6] J. Schmidhuber, D. Wierstra, M. Gagliolo, F. Gomez.
Training Recurrent Networks by Evolino. Neural Computation, 19(3): 757–779, 2007.
[7] H. Mayer, F. Gomez, D. Wierstra, I. Nagy, A. Knoll, and
J. Schmidhuber. A System for Robotic Heart Surgery that
Learns to Tie Knots Using Recurrent Neural Networks.
Advanced Robotics, 22/13–14, pp. 1521–1537, 2008.
[8] J. Schmidhuber and D. Wierstra and F. J. Gomez.
Evolino: Hybrid Neuroevolution / Optimal Linear Search for Sequence Learning. Proceedings of the 19th International Joint Conference on Artiﬁcial Intelligence (IJCAI),
Edinburgh, pp. 853–858, 2005.
[9] A. Graves and J. Schmidhuber. Framewise phoneme classiﬁcation with bidirectional LSTM and other neural network architectures. Neural Networks 18:5–6, pp. 602–
610, 2005.
[10] S. Fernandez, A. Graves, J. Schmidhuber. An application of recurrent neural networks to discriminative keyword spotting. Intl. Conf. on Artiﬁcial Neural Networks
ICANN'07, 2007.
[11] Graves, Alex; Mohamed, Abdel-rahman; Hinton, Geoffrey (2013). “Speech Recognition with Deep Recurrent
Neural Networks”. Acoustics, Speech and Signal Processing (ICASSP), 2013 IEEE International Conference on:
6645–6649.
[12] F. Gers, N. Schraudolph, J. Schmidhuber. Learning precise timing with LSTM recurrent networks. Journal of
Machine Learning Research 3:115–143, 2002.
[13] D. Eck and J. Schmidhuber. Learning The Long-Term
Structure of the Blues. In J. Dorronsoro, ed., Proceedings of Int. Conf. on Artiﬁcial Neural Networks ICANN'02,
Madrid, pages 284–289, Springer, Berlin, 2002.

12.6. EXTERNAL LINKS

[14] J. Schmidhuber, F. Gers, D. Eck. J. Schmidhuber, F.
Gers, D. Eck. Learning nonregular languages: A comparison of simple recurrent networks and LSTM. Neural
Computation 14(9):2039–2041, 2002.
[15] F. A. Gers and J. Schmidhuber. LSTM Recurrent Networks Learn Simple Context Free and Context Sensitive Languages. IEEE Transactions on Neural Networks
12(6):1333–1340, 2001.
[16] J. A. Perez-Ortiz, F. A. Gers, D. Eck, J. Schmidhuber.
Kalman ﬁlters improve LSTM network performance in problems unsolvable by traditional recurrent nets. Neural Networks 16(2):241–250, 2003.
[17] A. Graves, J. Schmidhuber. Oﬄine Handwriting Recognition with Multidimensional Recurrent Neural Networks.
Advances in Neural Information Processing Systems 22,
NIPS'22, pp 545–552, Vancouver, MIT Press, 2009.
[18] A. Graves, S. Fernandez,M. Liwicki, H. Bunke, J.
Schmidhuber. Unconstrained online handwriting recognition with recurrent neural networks. Advances in Neural Information Processing Systems 21, NIPS'21, pp 577–
584, 2008, MIT Press, Cambridge, MA, 2008.
[19] M. Baccouche, F. Mamalet, C Wolf, C. Garcia, A.
Baskurt. Sequential Deep Learning for Human Action
Recognition. 2nd International Workshop on Human Behavior Understanding (HBU), A.A. Salah, B. Lepri ed.
Amsterdam, Netherlands. pp. 29–39. Lecture Notes in
Computer Science 7065. Springer. 2011
[20] Hochreiter, S.; Heusel, M.; Obermayer, K. (2007).
“Fast model-based protein homology detection without alignment”. Bioinformatics 23 (14): 1728–1736. doi:10.1093/bioinformatics/btm247. PMID 17488755.

12.6 External links
• Recurrent Neural Networks with over 30 LSTM papers by Jürgen Schmidhuber's group at IDSIA
• Gers PhD thesis on LSTM networks.
• Fraud detection paper with two chapters devoted to explaining recurrent neural networks, especially
LSTM.
• Paper on a high-performing extension of LSTM that has been simpliﬁed to a single node type and can train arbitrary architectures.
• Tutorial: How to implement LSTM in python with theano 67

Chapter 13

Google Brain
Google Brain is a deep learning research project at
Google.

13.2 In Google products
The project's technology is currently used in the Android
Operating System's speech recognition system* [18] and photosearch for Google+.* [19]

13.1 History

Stanford University professor Andrew Ng who, since around 2006, became interested in using deep learning 13.3 Team techniques to crack the problem of artiﬁcial intelligence, started a deep learning project in 2011 as one of the
Google Brain was initially established by Google FelGoogle X projects. He then joined with Google Fellow low Jeﬀ Dean and visiting Stanford professor Andrew
Jeﬀ Dean to develop “the Google brain”as it was ﬁrst
Ng* [20] (Ng since moved to lead the artiﬁcial intelligence
*
referred to. [1] group at Baidu* [21]). As of 2014, team members inIn June 2012, the New York Times reported that a clus- cluded Jeﬀ Dean, Geoﬀrey Hinton, Greg Corrado, Quoc ter of 16,000 computers dedicated to mimicking some Le,* [22] Ilya Sutskever, Alex Krizhevsky, Samy Bengio, aspects of human brain activity had successfully trained and Vincent Vanhoucke. itself to recognize a cat based on 10 million digital images taken from YouTube videos.* [1] The story was also covered by National Public Radio* [2] and SmartPlanet.* [3]

13.4 Reception

In March 2013, Google hired Geoﬀrey Hinton, a leading researcher in the deep learning ﬁeld, and acquired the
Google Brain has received in-depth coverage in Wired company DNNResearch Inc. headed by Hinton. Hinton
Magazine,* [6]* [16]* [23] the New York Times,* [23] said that he would be dividing his future time between his
Technology Review,* [5]* [17] National Public Radio,* [2] university research and his work at Google.* [4] and Big Think.* [24]
On 26 January 2014, multiple news outlets stated that Google had purchased DeepMind Technologies for an undisclosed amount. Analysts later announced that the company was purchased for £400 Million 13.5 See also
($650M USD / €486M), although later reports estimated the acquisition was valued at over £500 Mil• Google X lion.* [5]* [6]* [7]* [8]* [9]* [10]* [11] The acquisition re• Google Research portedly took place after Facebook ended negotiations with DeepMind Technologies in 2013, which resulted in
• Quantum Artiﬁcial Intelligence Lab run by Google no agreement or purchase of the company.* [12] Google in collaboration with NASA and Universities Space has yet to comment or make an oﬃcial announcement on
Research Association. this acquisition.
In December 2012, futurist and inventor Ray Kurzweil, author of The Singularity is Near, joined Google in a full-time engineering director role, but focusing on the deep learning project.* [13] It was reported that Kurzweil would have“unlimited resources” pursue his vision at to Google.* [14]* [15]* [16]* [17] However, he is leading his own team, which is independent of Google Brain.

13.6 References

68

[1] Markoﬀ, John (June 25, 2012). “How Many Computers to Identify a Cat? 16,000”. New York Times. Retrieved
February 11, 2014.

13.6. REFERENCES

69

[2] “A Massive Google Network Learns To Identify —Cats” [19] “Improving Photo Search: A Step Across the Semantic
. National Public Radio. June 26, 2012. Retrieved FebruGap”. Google Research Blog. Google. June 12, 2013. ary 11, 2014.
[20] Jeﬀ Dean and Andrew Ng (26 June 2012). “Using largescale brain simulations for machine learning and A.I.”.
[3] Shin, Laura (June 26, 2012). “Google brain simulator
Oﬃcial Google Blog. Retrieved 26 January 2015. teaches itself to recognize cats”. SmartPlanet. Retrieved
February 11, 2014.
[21] “Ex-Google Brain head Andrew Ng to lead Baidu's artiﬁcial intelligence drive”. South China Morning Post.
[4] “U of T neural networks start-up acquired by Google”
(Press release). Toronto, ON. 12 March 2013. Retrieved
[22] “Quoc Le - Behind the Scenes”. Retrieved 20 April
13 March 2013.
2015.
[5] Regalado, Antonio (January 29, 2014). “Is Google Cornering the Market on Deep Learning? A cutting-edge cor- [23] Hernandez, Daniela (May 7, 2013). “The Man Behind the Google Brain: Andrew Ng and the Quest for the New ner of science is being wooed by Silicon Valley, to the
AI”. Wired Magazine. Retrieved February 11, 2014. dismay of some academics.”. Technology Review. Retrieved February 11, 2014.
[6] Wohlsen, Marcus (January 27, 2014).“Google’ Grand s Plan to Make Your Brain Irrelevant”. Wired Magazine.
Retrieved February 11, 2014.
[7] “Google Acquires UK AI startup Deepmind”. The
Guardian. Retrieved 27 January 2014.
[8] “Report of Acquisition, TechCrunch” TechCrunch. Re. trieved 27 January 2014.
[9] Oreskovic, Alexei.“Reuters Report” Reuters. Retrieved
.
27 January 2014.
[10] “Google Acquires Artiﬁcial Intelligence Start-Up DeepMind”. The Verge. Retrieved 27 January 2014.
[11] “Google acquires AI pioneer DeepMind Technologies”.
Ars Technica. Retrieved 27 January 2014.
[12] “Google beats Facebook for Acquisition of DeepMind
Technologies”. Retrieved 27 January 2014.
[13] Taylor, Colleen (December 14, 2012). “Ray Kurzweil
Joins Google In Full-Time Engineering Director Role;
Will Focus On Machine Learning, Language Processing”
. TechCrunch. Retrieved February 11, 2014.
[14] Empson, Rip (January 3, 2013). “Imagining The Future: Ray Kurzweil Has “Unlimited Resources”For AI,
Language Research At Google”. TechCrunch. Retrieved
February 11, 2014.
[15] Ferenstein, Gregory (January 6, 2013). “Google’s New
Director Of Engineering, Ray Kurzweil, Is Building Your
‘Cybernetic Friend’". TechCrunch. Retrieved February
11, 2014.
[16] Levy, Steven (April 25, 2013). “How Ray Kurzweil Will
Help Google Make the Ultimate AI Brain”. Wired Magazine. Retrieved February 11, 2014.
[17] Hof, Robert (April 23, 2013). “Deep Learning:
With massive amounts of computational power, machines can now recognize objects and translate speech in real time. Artiﬁcial intelligence is ﬁnally getting smart.”.
Technology Review. Retrieved February 11, 2014.
[18] “Speech Recognition and Deep Learning”. Google Research Blog. Google. August 6, 2012. Retrieved February
11, 2014.

[24] “Ray Kurzweil and the Brains Behind the Google Brain”
. Big Think. December 8, 2013. Retrieved February 11,
2014.

Chapter 14

Google DeepMind
Google DeepMind is a British artiﬁcial intelligence com- 14.2 Research pany. Founded in 2011 as DeepMind Technologies, it was acquired by Google in 2014.
DeepMind Technologies's goal is to "solve intelligence",* [21] which they are trying to achieve by combining "the best techniques from machine learning and systems neuroscience to build powerful general-purpose learning algorithms". * [21] They are trying to formalize
14.1 History intelligence* [22] in order to not only implement it into machines, but also understand the human brain, as Demis
14.1.1 2011 to 2014
Hassabis explains:
In 2011 the start-up was founded by Demis Hassabis,
Shane Legg and Mustafa Suleyman.* [3]* [4] Hassabis and
Legg ﬁrst met at UCL's Gatsby Computational Neuroscience Unit.* [5]

[...] Attempting to distil intelligence into an algorithmic construct may prove to be the best path to understanding some of the enduring mysteries of our minds.
—Demis Hassabis, Nature (journal), 23
February 2012* [23]

Since then major venture capitalist ﬁrms Horizons Ventures and Founders Fund have invested in the company,* [6] as well as entrepreneur Scott Banister.* [7] Jaan
Tallinn was an early investor and an advisor to the comCurrently the company's focus is on publishing research pany.* [8] on computer systems that are able to play games, and
In 2014, DeepMind received the“Company of the Year” developing these systems, ranging from strategy games
*
award by Cambridge Computer Laboratory. [9] such as Go* [24] to arcade games. According to Shane
The company has created a neural network that learns Legg human-level machine intelligence can be achieved how to play video games in a similar fashion to hu- "when a machine can learn to play a really wide range of mans* [10] and a neural network that may be able to ac- games from perceptual stream input and output, and transcess an external memory like a conventional Turing ma- fer understanding across games[...]."* [25] Research dechine, resulting in a computer that appears to possibly scribing an AI playing seven diﬀerent Atari video games mimic the short-term memory of the human brain.* [11] (Pong, Breakout, Space Invaders, Seaquest, Beamrider,
Enduro, and Q*bert) reportedly led to their acquisition by Google.* [10]

14.1.2

Acquisition by Google
14.2.1 Deep reinforcement learning
*

On 26 January 2014, Google announced [12] that it had agreed to acquire DeepMind Technologies. The acquisition reportedly took place after Facebook ended negotiations with DeepMind Technologies in 2013.* [13] Following the acquisition, the company was renamed Google
DeepMind.* [1]

As opposed to other AI's, such as IBM's Deep Blue or
Watson, which were developed for a pre-deﬁned purpose and only function within its merit, DeepMind claims that their system is not pre-programmed: it learns from experience, using only raw pixels as data input.* [1]* [26]
They test the system on video games, notably early arcade
Estimates of the cost of acquisition vary, from $400 mil- games, such as Space Invaders or Breakout.* [26]* [27] lion* [14] to over £500 million.* [15]* [16]* [17]* [18]* [19] Without altering the code, the AI begins to understand
One of DeepMind's conditions for Google was that they how to play the game, and after some time plays, for a few establish an AI Ethics committee.* [20] games (most notably Breakout), a more eﬃcient game
70

14.4. EXTERNAL LINKS

71

than any human ever could.* [27] For most games though [17] Oreskovic, Alexei.“Reuters Report” Reuters. Retrieved
.
27 January 2014.
(Space Invaders, Ms Pacman, Q*Bert for example),
DeepMind plays well below the current World Record.
[18] “Google Acquires Artiﬁcial Intelligence Start-Up DeepThe application of DeepMind's AI to video games is curMind”. The Verge. Retrieved 27 January 2014. rently for games made in the 1970s and 1980s, with work being done on more complex 3D games such as Doom, [19] “Google acquires AI pioneer DeepMind Technologies”.
Ars Technica. Retrieved 27 January 2014. which ﬁrst appeared in the early 1990s.* [27]

14.3 References
[1] Mnih1, Volodymyr; Kavukcuoglu1, Koray; Silver,
David (26 February 2015).
“Human-level control through deep reinforcement learning”. Nature. doi:10.1038/nature14236. Retrieved 25 February 2015.

[20] “Inside Google's Mysterious Ethics Board”. Forbes. 3
February 2014. Retrieved 12 October 2014.
[21] “DeepMind Technologies Website”. DeepMind Technologies. Retrieved 11 October 2014.
[22] Shane Legg; Joel Veness (29 September 2011). “An
Approximation of the Universal Intelligence Measure”
(PDF). Retrieved 12 October 2014.

[2] “Forbes Report - Acquisition”. Forbes. Retrieved 27
January 2014.

[23] Demis Hassabis (23 February 2012).“Model the brain’ s algorithms”(PDF). Nature. Retrieved 12 October 2014.

[3] “Google Buys U.K. Artiﬁcial Intelligence Company
DeepMind”. Bloomberg. 27 January 2014. Retrieved
13 November 2014.

[24] Shih-Chieh Huang; Martin Müller (12 July 2014).“Investigating the Limits of Monte-Carlo Tree Search Methods in Computer Go”. Springer.

[4] “Google makes £400m move in quest for artiﬁcial intelligence”. Financial Times. 27 January 2014. Retrieved
13 November 2014.

[25] “Q&A with Shane Legg on risks from AI”. 17 June
2011. Retrieved 12 October 2014.

[5] “Demis Hassabis: 15 facts about the DeepMind Technologies founder”. The Guardian. Retrieved 12 October
2014.

[26] Volodymyr Mnih; Koray Kavukcuoglu; David Silver;
Alex Graves; Ioannis Antonoglou; Daan Wierstra; Martin Riedmiller (12 December 2013). “Playing Atari with
Deep Reinforcement Learning”(PDF). Retrieved 12 October 2014.

[6] “DeepMind buy heralds rise of the machines” Financial
.
Times. Retrieved 14 October 2014.
[7] “DeepMind Technologies Investors”. Retrieved 12 October 2014.
[8] “Recode.net - DeepMind Technologies Acquisition”.
Retrieved 27 January 2014.
[9] “Hall of Fame Awards: To celebrate the success of companies founded by Computer Laboratory graduates.”.
Cambridge University. Retrieved 12 October 2014.
[10] “The Last AI Breakthrough DeepMind Made Before
Google Bought It”. The Physics arXiv Blog. Retrieved
12 October 2014.
[11] Best of 2014: Google's Secretive DeepMind Startup Unveils a
“Neural Turing Machine”MIT Technology Review
,
[12] “Google to buy artiﬁcial intelligence company DeepMind” Reuters. 26 January 2014. Retrieved 12 October
.
2014.
[13] “Google beats Facebook for Acquisition of DeepMind
Technologies”. Retrieved 27 January 2014.
[14] “Computers, gaming”. The Economist. 28 February
2015.
[15] “Google Acquires UK AI startup Deepmind”. The
Guardian. Retrieved 27 January 2014.
[16] “Report of Acquisition, TechCrunch” TechCrunch. Re. trieved 27 January 2014.

[27] Deepmind artiﬁcial intelligence @ FDOT14. 19 April
2014.

14.4 External links
• Google DeepMind

Chapter 15

Torch (machine learning)
Torch is an open source machine learning library, a scientiﬁc computing framework, and a script language based on the Lua programming language.* [3] It provides a wide range of algorithms for deep machine learning, and uses an extremely fast scripting language LuaJIT, and an underlying C implementation.

alized, as long as they do not contain references to objects that cannot be serialized, such as Lua coroutines, and Lua userdata. However, userdata can be serialized if it is wrapped by a table (or metatable) that provides read() and write() methods.

15.2 nn

15.1 torch
The core package of Torch is torch. It provides a ﬂexible N-dimensional array or Tensor, which supports basic routines for indexing, slicing, transposing, type-casting, resizing, sharing storage and cloning. This object is used by most other packages and thus forms the core object of the library. The Tensor also supports mathematical operations like max, min, sum, statistical distributions like uniform, normal and multinomial, and BLAS operations like dot product, matrix-vector multiplication, matrixmatrix multiplication, matrix-vector product and matrix product. The nn package is used for building neural networks.
It is divided into modular objects that share a common Module interface. Modules have a forward() and backward() method that allow them to feedforward and backpropagate, respectively. Modules can be joined together using module composites, like Sequential, Parallel and Concat to create complex task-tailored graphs. Simpler modules like Linear, Tanh and Max make up the basic component modules. This modular interface provides ﬁrst-order automatic gradient diﬀerentiation. What follows is an example use-case for building a multilayer perceptron using Modules:

The following exempliﬁes using torch via its REPL inter- > mlp = nn.Sequential() > mlp:add( nn.Linear(10, preter: 25) ) -- 10 input, 25 hidden units > mlp:add(
> a = torch.randn(3,4) > =a −0.2381 −0.3401 −1.7844 nn.Tanh() ) -- some hyperbolic tangent transfer
−0.2615 0.1411 1.6249 0.1708 0.8299 −1.0434 2.2291 function > mlp:add( nn.Linear(25, 1) ) -- 1 output >
1.0525 0.8465 [torch.DoubleTensor of dimension =mlp:forward(torch.randn(10)) −0.1815 [torch.Tensor
3x4] > a[1][2] −0.34010116549482 > a:narrow(1,1,2) of dimension 1]
−0.2381 −0.3401 −1.7844 −0.2615 0.1411 1.6249
0.1708 0.8299 [torch.DoubleTensor of dimension 2x4]
Loss functions are implemented as sub-classes of Crite> a:index(1, torch.LongTensor{1,2}) −0.2381 −0.3401 rion, which has a similar interface to Module. It also has
−1.7844 −0.2615 0.1411 1.6249 0.1708 0.8299 forward() and backward methods for computing the loss
[torch.DoubleTensor of dimension 2x4] > a:min() and backpropagating gradients, respectively. Criteria are
−1.7844365427828
helpful to train neural network on classical tasks. ComThe torch package also simpliﬁes object oriented programming and serialization by providing various convenience functions which are used throughout its packages. The torch.class(classname, parentclass) function can be used to create object factories (classes). When the constructor is called, torch initializes and sets a Lua table with the user-deﬁned metatable, which makes the table an object.

mon criteria are the Mean Squared Error criterion implemented in MSECriterion and the cross-entropy criterion implemented in ClassNLLCriterion. What follows is an example of a Lua function that can be iteratively called to train an mlp Module on input Tensor x, target Tensor y with a scalar learningRate:

function gradUpdate(mlp,x,y,learningRate) local criterion = nn.ClassNLLCriterion() pred = mlp:forward(x) local err = criterion:forward(pred,
Objects created with the torch factory can also be seri- y); mlp:zeroGradParameters(); local t =
72

15.6. EXTERNAL LINKS criterion:backward(pred, y); mlp:backward(x, mlp:updateParameters(learningRate); end

73
t);

15.6 External links
• Oﬃcial website

It also has StochasticGradient class for training a neural network using Stochastic gradient descent, although the
Optim package provides much more options in this respect, like momentum and weight decay regularization.

15.3 Other packages
Many packages other than the above oﬃcial packages are used with Torch. These are listed in the torch cheatsheet. These extra packages provide a wide range of utilities such as parallelism, asynchronous input/output, image processing, and so on.

15.4 Applications
Torch is used by Google DeepMind,* [4] the Facebook AI
Research Group,* [5] IBM,* [6] Yandex* [7] and the Idiap
Research Institute.* [8] Torch has been extended for use on Android* [9] and iOS.* [10] It has been used to build hardware implementations for data ﬂows like those found in neural networks.* [11]
Facebook has released a set of extension modules as open source software.* [12]

15.5 References
[1] “Torch: a modular machine learning software library”.
30 October 2002. Retrieved 24 April 2014.
[2] Ronan Collobert. “Torch7”. GitHub.
[3] Ronan Collobert; Koray Kavukcuoglu; Clement Farabet
(2011). “Torch7: A Matlab-like Environment for Machine Learning” (PDF). Neural Information Processing
Systems.
[4] What is going on with DeepMind and Google?
[5] KDnuggets Interview with Yann LeCun, Deep Learning
Expert, Director of Facebook AI Lab
[6] Hacker News
[7] Yann Lecun's Facebook Page
[8] IDIAP Research Institute : Torch
[9] Torch-android GitHub repository
[10] Torch-ios GitHub repository
[11] NeuFlow: A Runtime Reconﬁgurable Dataﬂow Processor for Vision
[12] “Facebook Open-Sources a Trove of AI Tools”. Wired.
16 January 2015.

• https://github.com/torch/torch7

Chapter 16

Theano (software)
Theano is a numerical computation library for
Python.* [1] In Theano, computations are expressed using a NumPy-like syntax and compiled to run eﬃciently on either CPU or GPU architectures.
Theano is an open source project* [2] primarily developed by a machine learning group at the Université de Montréal.* [3]

16.1 See also
• SciPy
• Torch

16.2 References
[1] Bergstra, J.; O. Breuleux, F. Bastien, P. Lamblin, R. Pascanu, G. Desjardins, J. Turian, D. Warde-Farley and Y.
Bengio (30 June 2010).“Theano: A CPU and GPU Math
Expression Compiler” (PDF). Proceedings of the Python for Scientiﬁc Computing Conference (SciPy) 2010.
[2] “Github Repository”.
[3] “deeplearning.net”.

74

Chapter 17

Deeplearning4j
Deeplearning4j is an open source deep learning library written for Java and the Java Virtual Machine* [1]* [2] and a computing framework with wide support for deep learning algorithms. Deeplearning4j includes implementations of the restricted Boltzmann machine, deep belief net, deep autoencoder, stacked denoising autoencoder and recursive neural tensor network, as well as word2vec, doc2vec and GloVe. These algorithms all include distributed parallel versions that integrate with
Hadoop and Spark.* [3]

17.3 Scientiﬁc Computing for the
JVM

17.1 Introduction

17.4 Canova Vectorization Lib for
Machine-Learning

Deeplearning4j includes an n-dimensional array class using ND4J that allows for scientiﬁc computing in Java and
Scala, similar to the functionality that Numpy provides to Python. It's eﬀectively based on a library for linear algebra and matrix manipulation in a production environment. It relies on Matplotlib as a plotting package.

Deeplearning4j relies on the widely used programming language, Java - though it is compatible with Clojure and includes a Scala API. It is powered by its own open-source Canova vectorizes various ﬁle formats and data types usnumerical computing library, ND4J, and works with both ing an input/output format system similar to Hadoop's use of MapReduce. A work in progress, Canova is designed
CPUs and GPUs.* [4] * [5] to vectorize CSVs, images, sound, text and video. Since
Deeplearning4j is an open source project* [6] primarily vectorization is a necessary step in preparing data to be developed by a machine learning group in San Francisco ingested by neural nets, Canova solves one of the most led by Adam Gibson.* [7]* [8] Deeplearning4j is the only important problems in machine learning. Canova can be open-source project listed on Google's Word2vec page used from the command line. for its Java implementation.* [9]
Deeplearning4j has been used in a number of commercial and academic applications. The code is hosted on GitHub* [10] and a support forum is maintained on
Google Groups.* [11]

17.5 Text & NLP

Deeplearning4j includes a vector space modeling and
The framework is composable, meaning shallow neural topic modeling toolkit, implemented in Java and integratnets such as restricted Boltzmann machines, convolu- ing with parallel GPUs for performance. It is speciﬁcally tional nets, autoencoders and recurrent nets can be added intended for handling large text collections. to one another to create deep nets of varying types.
Deeplearning4j includes implementations of tf–idf, deep learning, and Mikolov's word2vec algorithm, doc2vec and GloVe -- reimplemented and optimized in Java. It
17.2 Distributed relies on TSNE for word-cloud visualizations.
Training with Deeplearning4j takes place in the cluster, which means it can process massive amounts of data.
Neural nets are trained in parallel via iterative reduce, which works on Hadoop/YARN and on Spark.* [7]* [12]
Deeplearning4j also integrates with Cuda kernels to conduct pure GPU operations, and works with distributed
GPUs.

17.6 See also

75

• Torch
• Theano

76

17.7 References
[1] Metz, Cade (2014-06-02). “The Mission to Bring
Google's AI to the Rest of the World”. Wired.com. Retrieved 2014-06-28.
[2] Vance, Ashlee (2014-06-03).“Deep Learning for (Some of) the People”. Bloomberg Businessweek. Retrieved
2014-06-28.
[3] TV, Functional (2015-02-12).
“Adam Gibson,
DeepLearning4j on Spark and Data Science on JVM with nd4j, SF Spark @Galvanize 20150212”. SF Spark
Meetup. Retrieved 2015-03-01.
[4] Harris, Derrick (2014-06-02). “A startup called Skymind launches, pushing open source deep learning”.
GigaOM.com. Retrieved 2014-06-29.
[5] Novat, Jordan (2014-06-02). “Skymind launches with open-source, plug-and-play deep learning features for your app”. Retrieved 2014-06-29.
[6] “Github Repository”.
[7] “deeplearning4j.org”.
[8] “Crunchbase Proﬁle”.
[9] “Google Code”.
[10] Deeplearning4j source code
[11] Deeplearning4j Google Group
[12] “Iterative reduce”.

17.8 External links
• Oﬃcial website
• “Github Repositories”.
• “Deeplearning4j vs. Torch vs. Caﬀe vs. Pylearn”
.
• “Canova: A General Vectorization Lib for Machine
Learning”.
• “Apache Flink”.

CHAPTER 17. DEEPLEARNING4J

Chapter 18

Gensim
Gensim is an open-source vector space modeling and topic modeling toolkit, implemented in the Python programming language, using NumPy, SciPy and optionally
Cython for performance. It is speciﬁcally intended for handling large text collections, using eﬃcient online algorithms.

[8] Rehurek, Radim.“Gensim” http://radimrehurek.com/''.
.
Retrieved 27 January 2015. Gensim's tagline: "Topic
Modelling for Humans"

18.3 External links

Gensim includes implementations of tf–idf, random projections, deep learning with Google's word2vec algorithm * [1] (reimplemented and optimized in Cython), hierarchical Dirichlet processes (HDP), latent semantic analysis (LSA) and latent Dirichlet allocation (LDA), including distributed parallel versions.* [2]
Gensim has been used in a number of commercial as well as academic applications.* [3]* [4] The code is hosted on
GitHub* [5] and a support forum is maintained on Google
Groups.* [6]
Gensim accompanied the PhD dissertation Scalability of Semantic Analysis in Natural Language Processing of
Radim Řehůřek (2011).* [7]

18.1 Gensim's tagline
• Topic Modelling for Humans * [8]

18.2 References
[1] Deep learning with word2vec and gensim
[2] Radim Řehůřek and Petr Sojka (2010). Software framework for topic modelling with large corpora. Proc. LREC
Workshop on New Challenges for NLP Frameworks.
[3] Interview with Radim Řehůřek, creator of gensim
[4] gensim academic citations
[5] gensim source code
[6] gensim mailing list
[7] Rehurek, Radim (2011). “Scalability of Semantic Analysis in Natural Language Processing” (PDF). http:// radimrehurek.com/''. Retrieved 27 January 2015. my open-source gensim software package that accompanies this thesis

77

• Oﬃcial website

Chapter 19

Geoﬀrey Hinton
Geoﬀrey (Geoﬀ) Everest Hinton FRS (born 6 December 1947) is a British-born cognitive psychologist and computer scientist, most noted for his work on artiﬁcial neural networks. He now divides his time working for
Google and University of Toronto.* [1] He is the coinventor of the backpropagation and contrastive divergence training algorithms and is an important ﬁgure in the deep learning movement.* [2]

invented Boltzmann machines with Terry Sejnowski. His other contributions to neural network research include distributed representations, time delay neural network, mixtures of experts, Helmholtz machines and Product of Experts. His current main interest is in unsupervised learning procedures for neural networks with rich sensory input. 19.3 Honours and awards

19.1 Career
Hinton graduated from Cambridge in 1970, with a
Bachelor of Arts in experimental psychology, and from
Edinburgh in 1978, with a PhD in artiﬁcial intelligence. He has worked at Sussex, University of California San Diego, Cambridge, Carnegie Mellon University and University College London. He was the founding director of the Gatsby Computational Neuroscience Unit at
University College London, and is currently a professor in the computer science department at the University of
Toronto. He holds a Canada Research Chair in Machine
Learning. He is the director of the program on “Neural
Computation and Adaptive Perception”which is funded by the Canadian Institute for Advanced Research. Hinton taught a free online course on Neural Networks on the education platform Coursera in 2012.* [3] Hinton joined
Google in March 2013 when his company, DNNresearch
Inc, was acquired. He is planning to “divide his time between his university research and his work at Google”
.* [4]

19.2 Research interests
An accessible introduction to Geoﬀrey Hinton's research can be found in his articles in Scientiﬁc American in
September 1992 and October 1993. He investigates ways of using neural networks for learning, memory, perception and symbol processing and has authored over
200 publications in these areas. He was one of the researchers who introduced the back-propagation algorithm for training multi-layer neural networks that has been widely used for practical applications. He co-

Hinton was the ﬁrst winner of the David E. Rumelhart
Prize. He was elected a Fellow of the Royal Society in
1998.* [5]
In 2001, Hinton was awarded an Honorary Doctorate from the University of Edinburgh.
Hinton was the 2005 recipient of the IJCAI Award for
Research Excellence lifetime-achievement award.
He has also been awarded the 2011 Herzberg Canada
Gold Medal for Science and Engineering.* [6]
In 2013, Hinton was awarded an Honorary Doctorate from the Université de Sherbrooke.

19.4 Personal life
Hinton is the great-great-grandson both of logician
George Boole whose work eventually became one of the foundations of modern computer science, and of surgeon and author James Hinton.* [7] His father is Howard Hinton.

19.5 References

78

[1] Daniela Hernandez (7 May 2013). “The Man Behind the
Google Brain: Andrew Ng and the Quest for the New AI”
. Wired. Retrieved 10 May 2013.
[2] “How a Toronto professor’s research revolutionized artiﬁcial intelligence”. Toronto Star, Kate Allen, Apr 17
2015

19.6. EXTERNAL LINKS

[3] https://www.coursera.org/course/neuralnets
[4] “U of T neural networks start-up acquired by Google”
(Press release). Toronto, ON. 12 March 2013. Retrieved
13 March 2013.
[5] “Fellows of the Royal Society” The Royal Society. Re. trieved 14 March 2013.
[6] “Artiﬁcial intelligence scientist gets M prize” CBC News.
.
14 February 2011.
[7] The Isaac Newton of logic

19.6 External links
• Geoﬀrey E. Hinton's Academic Genealogy
• Geoﬀrey E. Hinton's Publications in Reverse
Chronological Order
• Homepage (at UofT)
• “The Next Generation of Neural Networks” on
YouTube
• Gatsby Computational Neuroscience Unit (founding director) • Encyclopedia article on Boltzmann Machines written by Geoﬀrey Hinton for Scholarpedia

79

Chapter 20

Yann LeCun
Yann LeCun (born 1960) is a computer scientist with contributions in machine learning, computer vision, mobile robotics and computational neuroscience. He is well known for his work on optical character recognition and computer vision using convolutional neural networks (CNN), and is a founding father of convolutional nets.* [1]* [2] He is also one of the main creators of the
DjVu image compression technology (together with Léon
Bottou and Patrick Haﬀner). He co-developed the Lush programming language with Léon Bottou.

20.1 Life
Yann LeCun was born near Paris, France, in 1960.
He received a Diplôme d'Ingénieur from the Ecole Superieure d'Ingénieur en Electrotechnique et Electronique
(ESIEE), Paris in 1983, and a PhD in Computer Science from Université Pierre et Marie Curie in 1987 during which he proposed an early form of the back-propagation learning algorithm for neural networks.* [3] He was a postdoctoral research associate in Geoﬀrey Hinton's lab at the University of Toronto.
In 1988, he joined the Adaptive Systems Research Department at AT&T Bell Laboratories in Holmdel, New
Jersey, USA, where he developed a number of new machine learning methods, such as a biologically inspired model of image recognition called Convolutional Neural
Networks,* [4] the“Optimal Brain Damage”regularization methods,* [5] and the Graph Transformer Networks method (similar to conditional random ﬁeld), which he applied to handwriting recognition and OCR.* [6] The bank check recognition system that he helped develop was widely deployed by NCR and other companies, reading over 10% of all the checks in the US in the late 1990s and early 2000s.

After a brief tenure as a Fellow of the NEC Research
Institute (now NEC-Labs America) in Princeton, NJ, he joined New York University (NYU) in 2003, where he is
Silver Professor of Computer Science Neural Science at the Courant Institute of Mathematical Science and the
Center for Neural Science. He is also a professor at
Polytechnic Institute of New York University.* [8]* [9] At
NYU, he has worked primarily on Energy-Based Models for supervised and unsupervised learning,* [10] feature learning for object recognition in Computer Vision,* [11] and mobile robotics.* [12]
In 2012, he became the founding director of the NYU
Center for Data Science.* [13] On December 9, 2013, LeCun became the ﬁrst director of Facebook AI Research in
New York City.,* [14] and stepped down from the NYUCDS directorship in early 2014.
LeCun is the recipient of the 2014 IEEE Neural Network
Pioneer Award.
In 2013, he and Yoshua Bengio co-founded the International Conference on Learning Representations, which adopted a post-publication open review process he previously advocated on his website. He was the chair and organizer of the“Learning Workshop” held every year between 1986 and 2012 in Snowbird, Utah. He is a member of the Science Advisory Board of the Institute for Pure and Applied Mathematics* [15] at UCLA, and has been on the advisory board of a number of companies, including MuseAmi, KXEN Inc., and Vidient Systems.* [16]
He is the Co-Director of the Neural Computation &
Adaptive Perception research program of CIFAR* [17]

20.2 References

In 1996, he joined AT&T Labs-Research as head of the
Image Processing Research Department, which was part of Lawrence Rabiner's Speech and Image Processing Research Lab, and worked primarily on the DjVu image compression technology,* [7] used by many websites, notably the Internet Archive, to distribute scanned documents. His collaborators at AT&T include Léon Bottou and Vladimir Vapnik.
80

[1] Convolutional Nets and CIFAR-10: An Interview with
Yann LeCun. Kaggle 2014
[2] LeCun, Yann; Léon Bottou; Yoshua Bengio; Patrick
Haﬀner (1998).“Gradient-based learning applied to document recognition” (PDF). Proceedings of the IEEE 86
(11): 2278–2324. doi:10.1109/5.726791. Retrieved 16
November 2013.
[3] Y. LeCun: Une procédure d'apprentissage pour réseau a seuil asymmetrique (a Learning Scheme for Asymmetric

20.3. EXTERNAL LINKS

81

Threshold Networks), Proceedings of Cognitiva 85, 599–
604, Paris, France, 1985.
[4] Y. LeCun, B. Boser, J. S. Denker, D. Henderson, R. E.
Howard, W. Hubbard and L. D. Jackel: Backpropagation
Applied to Handwritten Zip Code Recognition, Neural
Computation, 1(4):541-551, Winter 1989.
[5] Yann LeCun, J. S. Denker, S. Solla, R. E. Howard and L.
D. Jackel: Optimal Brain Damage, in Touretzky, David
(Eds), Advances in Neural Information Processing Systems 2 (NIPS*89), Morgan Kaufmann, Denver, CO,
1990.
[6] Yann LeCun, Léon Bottou, Yoshua Bengio and Patrick
Haﬀner: Gradient Based Learning Applied to Document
Recognition, Proceedings of IEEE, 86(11):2278–2324,
1998.
[7] Léon Bottou, Patrick Haﬀner, Paul G. Howard, Patrice
Simard, Yoshua Bengio and Yann LeCun: High Quality Document Image Compression with DjVu, Journal of
Electronic Imaging, 7(3):410–425, 1998.
[8] “People - Electrical and Computer Engineering”. Polytechnic Institute of New York University. Retrieved 13
March 2013.
[9] http://yann.lecun.com/
[10] Yann LeCun, Sumit Chopra, Raia Hadsell, Ranzato
Marc'Aurelio and Fu-Jie Huang: A Tutorial on EnergyBased Learning, in Bakir, G. and Hofman, T. and
Schölkopf, B. and Smola, A. and Taskar, B. (Eds), Predicting Structured Data, MIT Press, 2006.
[11] Kevin Jarrett, Koray Kavukcuoglu, Marc'Aurelio Ranzato and Yann LeCun: What is the Best Multi-Stage Architecture for Object Recognition?, Proc. International Conference on Computer Vision (ICCV'09), IEEE, 2009
[12] Raia Hadsell, Pierre Sermanet, Marco Scoﬃer, Ayse
Erkan, Koray Kavackuoglu, Urs Muller and Yann LeCun: Learning Long-Range Vision for Autonomous OﬀRoad Driving, Journal of Field Robotics, 26(2):120–144,
February 2009.
[13] http://cds.nyu.edu
[14] https://www.facebook.com/yann.lecun/posts/
10151728212367143
[15] http://www.ipam.ucla.edu/programs/gss2012/ for Pure and Applied Mathematics

Institute

[16] Vidient Systems.
[17] “Neural Computation & Adaptive Perception Advisory
.
Committee Yann LeCun” CIFAR. Retrieved 16 December 2013.

20.3 External links
• Yann LeCun's personal website
• Yann LeCun's lab website at NYU

• Yann LeCun's List of PhD Students
• Yann LeCun's publications
• Convolutional Neural Networks
• DjVuLibre website
• Lush website

Chapter 21

Jürgen Schmidhuber
Jürgen Schmidhuber (born 17 January 1963 in
Munich) is a computer scientist and artist known for his work on machine learning, Artiﬁcial Intelligence
(AI), artiﬁcial neural networks, digital physics, and lowcomplexity art. His contributions also include generalizations of Kolmogorov complexity and the Speed
Prior. From 2004 to 2009 he was professor of Cognitive Robotics at the Technische Universität München.
Since 1995 he has been co-director of the Swiss AI
Lab IDSIA in Lugano, since 2009 also professor of Artiﬁcial Intelligence at the University of Lugano. Between 2009 and 2012, the recurrent neural networks and deep feedforward neural networks developed in his research group have won eight international competitions in pattern recognition and machine learning.* [1] In honor of his achievements he was elected to the European
Academy of Sciences and Arts in 2008.

21.1 Contributions
21.1.1

Recurrent neural networks

The dynamic recurrent neural networks developed in his lab are simpliﬁed mathematical models of the biological neural networks found in human brains. A particularly successful model of this type is called Long short term memory.* [2] From training sequences it learns to solve numerous tasks unsolvable by previous such models. Applications range from automatic music composition to speech recognition, reinforcement learning and robotics in partially observable environments. As of 2010, his group has the best results on benchmarks in automatic handwriting recognition, obtained with deep neural networks* [3] and recurrent neural networks.* [4]

21.1.2

Artiﬁcial evolution / genetic programming

As an undergrad at TUM Schmidhuber evolved computer programs through genetic algorithms. The method was published in 1987 as one of the ﬁrst papers in the emerging ﬁeld that later became known as genetic pro-

gramming. In the same year he published the ﬁrst work on Meta-genetic programming. Since then he has co-authored numerous additional papers on artiﬁcial evolution. Applications include robot control, soccer learning, drag minimization, and time series prediction.
He received several best paper awards at scientiﬁc conferences on evolutionary computation.

21.1.3 Neural economy
In 1989 he created the ﬁrst learning algorithm for neural networks based on principles of the market economy (inspired by John Holland's bucket brigade algorithm for classiﬁer systems): adaptive neurons compete for being active in response to certain input patterns; those that are active when there is external reward get stronger synapses, but active neurons have to pay those that activated them, by transferring parts of their synapse strengths, thus rewarding “hidden”neurons setting the stage for later success.* [5]

21.1.4 Artiﬁcial curiosity and creativity
In 1990 he published the ﬁrst in a long series of papers on artiﬁcial curiosity and creativity for an autonomous agent.
The agent is equipped with an adaptive predictor trying to predict future events from the history of previous events and actions. A reward-maximizing, reinforcement learning, adaptive controller is steering the agent and gets curiosity reward for executing action sequences that improve the predictor. This discourages it from executing actions leading to boring outcomes that are either predictable or totally unpredictable.* [6] Instead the controller is motivated to learn actions that help the predictor to learn new, previously unknown regularities in its environment, thus improving its model of the world, which in turn can greatly help to solve externally given tasks.
This has become an important concept of developmental robotics. Schmidhuber argues that his corresponding formal theory of creativity explains essential aspects of art, science, music, and humor.* [7]

82

21.2. REFERENCES

21.1.5

Unsupervised learning / factorial 21.1.8 Low-complexity art / theory of codes beauty

During the early 1990s Schmidhuber also invented a neural method for nonlinear independent component analysis (ICA) called predictability minimization. It is based on co-evolution of adaptive predictors and initially random, adaptive feature detectors processing input patterns from the environment. For each detector there is a predictor trying to predict its current value from the values of neighboring detectors, while each detector is simultaneously trying to become as unpredictable as possible.* [8] It can be shown that the best the detectors can do is to create a factorial code of the environment, that is, a code that conveys all the information about the inputs such that the code components are statistically independent, which is desirable for many pattern recognition applications. 21.1.6

Schmidhuber's low-complexity artworks (since 1997) can be described by very short computer programs containing very few bits of information, and reﬂect his formal theory of beauty* [15] based on the concepts of Kolmogorov complexity and minimum description length.
Schmidhuber writes that since age 15 or so his main scientiﬁc ambition has been to build an optimal scientist, then retire. First he wants to build a scientist better than himself (he quips that his colleagues claim that should be easy) who will then do the remaining work. He claims he
“cannot see any more eﬃcient way of using and multiplying the little creativity he's got”.

21.1.9 Robot learning
In recent years a robotics group with focus on intelligent

Kolmogorov complexity / computer- and learning robots, especially in the ﬁelds of swarm and generated universe humanoid robotics was established at his lab.* [16] The

In 1997 Schmidhuber published a paper based on Konrad
Zuse's assumption (1967) that the history of the universe is computable. He pointed out that the simplest explanation of the universe would be a very simple Turing machine programmed to systematically execute all possible programs computing all possible histories for all types of computable physical laws.* [9]* [10] He also pointed out that there is an optimally eﬃcient way of computing all computable universes based on Leonid Levin's universal search algorithm (1973). In 2000 he expanded this work by combining Ray Solomonoﬀ's theory of inductive inference with the assumption that quickly computable universes are more likely than others.* [11] This work on digital physics also led to limit-computable generalizations of algorithmic information or Kolmogorov complexity and the concept of Super Omegas, which are limit-computable numbers that are even more random (in a certain sense) than Gregory Chaitin's number of wisdom
Omega.* [12]

21.1.7

83

Universal AI

Important research topics of his group include universal learning algorithms and universal AI* [13]* [14] (see
Gödel machine). Contributions include the ﬁrst theoretically optimal decision makers living in environments obeying arbitrary unknown but computable probabilistic laws, and mathematically sound general problem solvers such as the remarkable asymptotically fastest algorithm for all well-deﬁned problems, by his former postdoc
Marcus Hutter. Based on the theoretical results obtained in the early 2000s, Schmidhuber is actively promoting the view that in the new millennium the ﬁeld of general AI has matured and become a real formal science.

lab is equipped with a variety of mobile and ﬂying robots and is one of the around 20 labs in the world owning an iCub humanoid robot. The group has applied a variety of machine learning algorithms, such as reinforcement learning and genetic programming, to improve adaptiveness and autonomy of robotic systems.
Recently his work on evolutionary robotics, with a focus on using genetic programming to evolve robotic skills, especially in robot vision have allowed for quick and robust object detection in humanoid robots.* [17]* [18]* [19] IDSIA's work with the iCub humanoid won the 2013 AAAI
Student Video competition.* [20]* [21]

21.2 References
[1] 2012 Kurzweil AI Interview with Jürgen Schmidhuber on the eight competitions won by his Deep Learning team
2009-2012
[2] S. Hochreiter and J. Schmidhuber. Long Short-Term
Memory. Neural Computation, 9(8):1735–1780, 1997.
[3] D. C. Ciresan, U. Meier, L. M. Gambardella, J. Schmidhuber. Deep Big Simple Neural Nets For Handwritten
Digit Recognition. Neural Computation 22(12): 32073220.
[4] A. Graves, M. Liwicki, S. Fernandez, R. Bertolami, H.
Bunke, J. Schmidhuber. A Novel Connectionist System for Improved Unconstrained Handwriting Recognition.
IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 31, no. 5, 2009.
[5] J. Schmidhuber. A local learning algorithm for dynamic feedforward and recurrent networks. Connection Science,
1(4):403–412, 1989

84

[6] J. Schmidhuber. Curious model-building control systems.
In Proc. International Joint Conference on Neural Networks, Singapore, volume 2, pages 1458–1463. IEEE,
1991
[7] J. Schmidhuber. Formal Theory of Creativity, Fun, and
Intrinsic Motivation (1990–2010). IEEE Transactions on
Autonomous Mental Development, 2(3):230–247, 2010.
[8] J. Schmidhuber. Learning factorial codes by predictability minimization. Neural Computation, 4(6):863–879, 1992
[9] J. Schmidhuber. A computer scientist's view of life, the universe, and everything. Foundations of Computer Science: Potential – Theory – Cognition, Lecture Notes in
Computer Science, pages 201–208, Springer, 1997
[10] Brian Greene, Chapter 10 of: The Hidden Reality: Parallel Universes and the Deep Laws of the Cosmos, Knopf,
2011
[11] J. Schmidhuber. The Speed Prior: A New Simplicity
Measure Yielding Near-Optimal Computable Predictions.
Proceedings of the 15th Annual Conference on Computational Learning Theory (COLT 2002), Sydney, Australia,
LNAI, 216–228, Springer, 2002
[12] J. Schmidhuber. Hierarchies of generalized Kolmogorov complexities and nonenumerable universal measures computable in the limit. International Journal of Foundations of Computer Science 13(4):587–612, 2002
[13] J. Schmidhuber. Ultimate Cognition à la Gödel. Cognitive Computation 1(2):177–193, 2009
[14] J. Schmidhuber. Optimal Ordered Problem Solver. Machine Learning, 54, 211–254, 2004
[15] J. Schmidhuber. Low-Complexity Art. Leonardo, Journal of the International Society for the Arts, Sciences, and
Technology, 30(2):97–103, MIT Press, 1997
[16] http://robotics.idsia.ch/ The IDSIA Robotics Lab
[17] J. Leitner, S. Harding, P. Chandrashekhariah, M. Frank,
A. Förster, J. Triesch and J. Schmidhuber. Learning Visual Object Detection and Localisation Using icVision.
Biologically Inspired Cognitive Architectures, Vol. 5,
2013.
[18] J. Leitner, S. Harding, M. Frank, A. Förster and J.
Schmidhuber. Humanoid Learns to Detect Its Own
Hands. IEEE Congress on Evolutionary Computing
(CEC), 2013.
[19] S. Harding, J. Leitner and J. Schmidhuber. Cartesian
Genetic Programming for Image Processing (CGP-IP).
Genetic Programming Theory and Practice X (Springer
Tract on Genetic and Evolutionary Computation). pp 3144. ISBN 978-1-4614-6845-5. Springer, Ann Arbor,
2013.
[20] http://www.aaaivideos.org/2013/ AAAI Video Competition 2013.
[21] M. Stollenga, L. Pape, M. Frank, J. Leitner, A. Förster and
J. Schmidhuber. Task-Relevant Roadmaps: A Framework for Humanoid Motion Planning. IROS, 2013.

CHAPTER 21. JÜRGEN SCHMIDHUBER

21.3 Sources
• Google Scholar: Numerous scientiﬁc articles referencing Schmidhuber's work
• Scholarpedia article on Universal Search, discussing
Schmidhuber's Speed Prior, Optimal Ordered Problem Solver, Gödel machine
• German article on Schmidhuber in CIO magazine:
“Der ideale Wissenschaftler”(the ideal scientist)
• Build An Optimal Scientist, Then Retire: Interview with J. Schmidhuber in H+ magazine, 2010
• Video of Schmidhuber's talk on artiﬁcial curiosity and creativity at the Singularity Summit 2009, NYC
• TV clip: Schmidhuber on computable universes on
Through the Wormhole with Morgan Freeman.
• On-going research on the iCub humanoid at the IDSIA Robotics Lab

21.4 External links
• Home page
• Publications
• Videos of Juergen Schmidhuber & the Swiss AI Lab
IDSIA

Chapter 22

Jeﬀ Dean (computer scientist)
Jeﬀrey Adgate “Jeﬀ”Dean (born 1968) is an American involvement in the engineering hiring process. computer scientist and software engineer. He is currently
Among others, the projects he's worked on include: a Google Senior Fellow in the Systems and Infrastructure
Group.
• Spanner - a scalable, multi-version, globally distributed, and synchronously replicated database

22.1 Personal life and education

• Some of the production system design and statistical machine translation system for Google Translate.

Dean received a Ph.D. in Computer Science from the
University of Washington, working with Craig Chambers on whole-program optimization techniques for objectoriented languages. He received a B.S., summa cum laude from the University of Minnesota in Computer Science
& Economics in 1990. He was elected to the National
Academy of Engineering in 2009, which recognized his work on“the science and engineering of large-scale distributed computer systems.”

• BigTable, a large-scale semi-structured storage system.
• MapReduce a system for large-scale data processing applications. • Google Brain a system for large-scale artiﬁcial neural networks

22.4 Awards and honors
22.2 Career in computer science

• Elected to the National Academy of Engineering
(2009)

Prior to joining Google, he was at DEC/Compaq's Western Research Laboratory, where he worked on proﬁling tools, microprocessor architecture, and information retrieval.
Prior to graduate school, he worked at the World Health
Organization's Global Programme on AIDS, developing software for statistical modeling and forecasting of the
HIV/AIDS pandemic.

• Fellow of the Association for Computing Machinery
(2009)
• ACM-Infosys Foundation Award (2012)

22.5 Major publications
• Jeﬀrey Dean and Sanjay Ghemawat.
2004.
MapReduce: Simpliﬁed Data Processing on Large
Clusters. OSDI'04: Sixth Symposium on Operating System Design and Implementation (December
2004)

22.3 Career at Google
Dean joined Google in mid-1999, and is currently a
Google Senior Fellow in the Systems Infrastructure
Group. While at Google, he has designed and implemented large portions of the company's advertising, crawling, indexing and query serving systems, along with various pieces of the distributed computing infrastructure that sits underneath most of Google's products. At various times, he has also worked on improving search quality, statistical machine translation, and various internal software development tools and has had signiﬁcant

22.6 See also

85

• Big Table
• MapReduce
• Spanner

86

22.7 External links
• Jeﬀ Dean's Google home page.
• Meet Google's Baddest Engineer, Jeﬀ Dean
• The Optimizer

CHAPTER 22. JEFF DEAN (COMPUTER SCIENTIST)

Chapter 23

Andrew Ng
Andrew Yan-Tak Ng (Chinese: 吳恩達; born 1976) is
Chief Scientist at Baidu Research in Silicon Valley. In addition, he is an associate professor in the Department of Computer Science and the Department of Electrical
Engineering by courtesy at Stanford University. He is chairman of the board of Coursera, an online education platform that he co-founded with Daphne Koller.

23.2 Online education

Ng is also the author or co-author of over 100 published papers in machine learning, robotics and related ﬁelds, and some of his work in computer vision has been featured in a series of press releases and reviews.* [5] In
2008, he was named to the MIT Technology Review TR35 as one of the top 35 innovators in the world under the age of 35.* [6]* [7] In 2007, Ng was awarded a Sloan Fellowship. For his work in Artiﬁcial Intelligence, he is also a recipient of the Computers and Thought Award.

led to the founding of Coursera in 2012.

Ng started the Stanford Engineering Everywhere (SEE) program, which in 2008 placed a number of Stanford courses online, for free. Ng taught one of these courses,
Machine Learning, which consisted of video lectures by him, along with the student materials used in the Stanford
He researches primarily in machine learning and deep CS229 class. learning. His early work includes the Stanford Au- The “applied”version of the Stanford class (CS229a) tonomous Helicopter project, which developed one was hosted on ml-class.org and started in October 2011, of the most capable autonomous helicopters in the with over 100,000 students registered for its ﬁrst iteration; world,* [2]* [3] and the STAIR (STanford Artiﬁcial In- the course featured quizzes and graded programming astelligence Robot) project,* [4] which resulted in ROS, a signments and became one of the ﬁrst successful MOOCs widely used open-source robotics software platform. made by Stanford professors.* [14] His work subsequently

23.3 Personal life

Ng was born in the UK in 1976. His parents were both Hongkongers. He spent time in Hong Kong and
Singapore* [1] and later graduated from Raﬄes Institution in Singapore as the class of 1992 and received his
On May 16, 2014, Ng announced from his Coursera blog undergraduate degree in computer science from Carnegie that he would be stepping away from his day-to-day re- Mellon University in Pittsburgh, Pennsylvania as the class sponsibilities at Coursera, and join Baidu as Chief Scien- of 1997. Then, he attained his master's degree from tist, working on deep learning.* [8]
Massachusetts Institute of Technology in Cambridge,
Massachusetts as the class of 1998 and received his PhD from University of California, Berkeley in 2002. He started working at Stanford University during that year; he currently lives in Palo Alto, California. He married
23.1 Machine learning research
Carol E. Reiley in 2014.
In 2011, Ng founded the Google Brain project at Google, which developed very large scale artiﬁcial neural networks using Google's distributed computer infrastructure.* [9] Among its notable results was a neural network trained using deep learning algorithms on 16,000 CPU cores, that learned to recognize higher-level concepts, such as cats, after watching only YouTube videos, and without ever having been told what a“cat” * [10]* [11] is. The project's technology is currently also used in the
Android Operating System's speech recognition system.* [12]

23.4 References

87

[1] Seligman, Katherine (3 December 2006). “If Andrew
Ng could just get his robot to assemble an Ikea bookshelf, we'd all buy one”. SFGate. Retrieved 12 February 2013.
[2] “From Self-Flying Helicopters to Classrooms of the Future”. Chronicle of Higher Education. 2012.
[3] “Stanford Autonomous Helicopter Project”.

88

[4] John Markoﬀ (18 July 2006).“Brainy Robots Start Stepping Into Daily Life”. New York Times.
[5] New algorithm improves robot vision
[6]“2008 Young Innovators Under 35” Technology Review.
.
2008. Retrieved August 15, 2011.
[7] Technology Review: TR35
[8] “A personal message from Co-founder Andrew Ng”.
Coursera blog. 2014. Retrieved May 16, 2014.
[9] Claire Miller and Nick Bilton (3 November 2011).
“Google’s Lab of Wildest Dreams”. New York Times.
[10] John Markoﬀ (25 June 2012). “How Many Computers to Identify a Cat? 16,000.”. New York Times.
[11] Ng, Andrew; Dean, Jeﬀ (2012). “Building High-level
Features Using Large Scale Unsupervised Learning”
(PDF).
[12] “Speech Recognition and Deep Learning”. Google Research Blog. Google. 6 August 2012. Retrieved 29 January 2013.
[13] “Interview with Coursera Co-Founder Andrew Ng” De. gree of Freedom. Retrieved May 19, 2013.
[14] Theresa Johnson.“Stanford for All” Stanford Magazine.
.

23.5 See also
• Robot Operating System

23.6 External links
• Homepage
• STAIR Homepage
• Publications
• Academic Genealogy
• Machine Learning (CS 229) Video Lecture
• Lecture videos
• From Self-Flying Helicopters to Classrooms of the
Future
• Coursera-Leadership

CHAPTER 23. ANDREW NG

23.7. TEXT AND IMAGE SOURCES, CONTRIBUTORS, AND LICENSES

89

23.7 Text and image sources, contributors, and licenses
23.7.1

Text

• Artiﬁcial neural network Source: http://en.wikipedia.org/wiki/Artificial%20neural%20network?oldid=662892335 Contributors: Magnus Manske, Ed Poor, Iwnbap, PierreAbbat, Youandme, Susano, Hfastedge, Mrwojo, Michael Hardy, Erik Zachte, Oliver Pereira,
Bobby D. Bryant, Zeno Gantner, Parmentier~enwiki, Delirium, Pieter Suurmond, (, Alﬁo, 168..., Ellywa, Ronz, Snoyes, Den fjättrade ankan~enwiki, Cgs, Glenn, Cyan, Hike395, Hashar, Novum, Charles Matthews, Guaka, Timwi, Reddi, Andrewman327, Munford, Furrykef, Bevo, Fvw, Raul654, Nyxos, Unknown, Pakcw, Robbot, Chopchopwhitey, Bkell, Hadal, Wikibot, Diberri, Xanzzibar, Wile E.
Heresiarch, Connelly, Giftlite, Rs2, Markus Krötzsch, Spazzm, Seabhcan, BenFrantzDale, Zigger, Everyking, Rpyle731, Wikiwikifast,
Foobar, Edrex, Jabowery, Wildt~enwiki, Wmahan, Neilc, Quadell, Beland, Lylum, Gene s, Sbledsoe, Mozzerati, Karl-Henner, Jmeppley, Asbestos, Fintor, AAAAA, Splatty, Rich Farmbrough, Pak21, NeuronExMachina, Michal Jurosz, Pjacobi, Mecanismo, Zarutian,
Dbachmann, Bender235, ZeroOne, Violetriga, Mavhc, One-dimensional Tangent, Gyll, Stephane.magnenat, Mysteronald, .:Ajvol:., Fotinakis, Nk, Tritium6, JesseHogan, Mdd, Passw0rd, Zachlipton, Alansohn, Jhertel, Anthony Appleyard, Denoir, Arthena, Fritz Saalfeld,
Sp00n17, Rickyp, Hu, Tyrell turing, Cburnett, Notjim, Drbreznjev, Forderud, Oleg Alexandrov, Mogigoma, Madmardigan53, Justinlebar, Olethros, Ylem, Dr.U, Gengiskanhg, Male1979, Bar0n, Waldir, Eslip17, Yoghurt, Ashmoo, Graham87, Qwertyus, Imersion, Grammarbot, Rjwilmsi, Jeema, Venullian, SpNeo, Intgr, Predictor, Kri, BradBeattie, Plarroy, Windharp, Mehran.asadi, Commander Nemet,
Wavelength, Borgx, IanManka, Rsrikanth05, Philopedia, Ritchy, David R. Ingham, Grafen, Nrets, Exir Kamalabadi, Deodar~enwiki,
Mosquitopsu, Jpbowen, Dennis!, JulesH, Moe Epsilon, Supten, DeadEyeArrow, Eclipsed, SamuelRiv, Tribaal, Chase me ladies, I'm the Cavalry, CWenger, Donhalcon, Banus, Shepard, John Broughton, A13ean, SmackBot, PinstripeMonkey, McGeddon, CommodiCast,
Jfmiller28, Stimpy, Commander Keane bot, Feshmania, ToddDeLuca, Diegotorquemada, Patrickdepinguin, KYN, Gilliam, Bluebot, Oli
Filth, Gardoma, Complexica, Nossac, Hongooi, Pdtl, Izhikevich, Trifon Triantaﬁllidis, SeanAhern, Neshatian, Vernedj, Dankonikolic,
Rory096, Sina2, SS2005, Kuru, Plison, Lakinekaki, Bjankuloski06en~enwiki, IronGargoyle, WMod-NS, Dicklyon, Citicat, StanfordProgrammer, Ojan, Chi3x10, Aeternus, CapitalR, Atreys, George100, Gveret Tered, Devourer09, SkyWalker, CmdrObot, Leonoel, CBM,
Mcstrother, MarsRover, CX, Arauzo, Peterdjones, Josephorourke, Kozuch, ClydeC, NotQuiteEXPComplete, Irigi, Mbell, Oldiowl, Tolstoy the Cat, Headbomb, Mitchell.E.Timin, Davidhorman, Sbandrews, KrakatoaKatie, QuiteUnusual, Prolog, AnAj, LinaMishima, Whenning,
Hamaryns, Daytona2, JAnDbot, MER-C, Dcooper, Extropian314, Magioladitis, VoABot II, Amitant, Jimjamjak, SSZ, Robotman1974,
David Eppstein, User A1, Martynas Patasius, Pmbhagat, JaGa, Tuhinsubhrakonar, SoyYo, Nikoladie~enwiki, R'n'B, Maproom, K.menin,
Gill110951, Tarotcards, Plasticup, Margareta, Paskari, Jamesontai, Kiran uvpce, Jamiejoseph, Error9312, Jlaramee, Jeﬀ G., A4bot, Singleheart, Ebbedc, Lordvolton, Ask123, CanOfWorms, Mundhenk, Wikiisawesome, M karamanov, Enkya, Blumenkraft, Twikir, Mikemoral, Oldag07, Smsarmad, Flyer22, Janopus, Bwieliczko, Dhatﬁeld, F.j.gaze, Mark Lewis Epstein, S2000magician, PuercoPop, Martarius, ClueBot, Ignacio Javier Igjav, Ahyeek, The Thing That Should Not Be, Fadesga, Zybler, Midiangr, Epsilon60198, Thomas Tvileren,
Wduch, Excirial, Three-quarter-ten, Skbkekas, Chaosdruid, Aprock, Qwfp, Jean-claude perez, Achler, XLinkBot, AgnosticPreachersKid,
BodhisattvaBot, Stickee, Cmr08, Porphyro, Fippy Darkpaw, Addbot, DOI bot, AndrewHZ, Thomblake, Techjerry, Looie496, MrOllie, Transmobilator, Jarble, Yobot, Blm19732008, Nguyengiap84~enwiki, SparkOfCreation, AnomieBOT, DemocraticLuntz, Tryptoﬁsh,
Trevithj, Jim1138, Durran65, MockDuck, JonathanWilliford, Materialscientist, Citation bot, Eumolpo, Twri, NFD9001, Isheden, J04n,
Omnipaedista, Mark Schierbecker, RibotBOT, RoodyBeep, Gunjan verma81, FrescoBot, X7q, Ömer Cengiz Çelebi, Outback the koala,
Citation bot 1, Tylor.Sampson, Calmer Waters, Skyerise, Trappist the monk, Krassotkin, Cjlim, Fox Wilson, The Strategist, LilyKitty,
Eparo, ‫ ,בן גרשון‬Jfmantis, Mehdiabbasi, VernoWhitney, Wiknn, BertSeghers, DASHBot, EmausBot, Nacopt, Dzkd, Racerx11, Japs 88,
GoingBatty, RaoInWiki, Roposeidon, Epsiloner, Stheodor, Benlansdell, Radshashi, K6ka, D'oh!, Thisisentchris87, Aavindraa, Chire,
Glosser.ca, IGeMiNix, Donner60, Yoshua.Bengio, Shinosin, Venkatarun95, ChuckNorrisPwnedYou, Petrb, ClueBot NG, Raghith, Robiminer, Snotbot, Tideﬂat, Frietjes, Gms3591, Ryansandersuk, Widr, MerlIwBot, Helpful Pixie Bot, Trepier, BG19bot, Thwien, Adams7,
Rahil2000, Michaelmalak, Compfreak7, Kirananils, Altaïr, J.Davis314, Attleboro, Pratyya Ghosh, JoshuSasori, Ferrarisailor, Eugenecheung, Mtschida, ChrisGualtieri, Dave2k6inthemix, Whebzy, APerson, JurgenNL, Oritnk, Stevebillings, Djfrost711, Sa publishers, 㓟, Mark viking, Markus.harz, Deeper Learning, Vinchaud20, Soueumxm, Toritris, Evolution and evolvability, Sboddhu, Sharva029, Paheld, Putting things straight, Rosario Berganza, Monkbot, Buggiehuggie, Santoshwriter, Likerhayter, Joma.huguet, Bclark401, Rahulpratapsingh06, Donkeychee, Michaelwine, Xsantostill, Jorge Guerra Pires, Wfwhitney, Loïc Bourgois, KasparBot and Anonymous: 488
• Deep learning Source: http://en.wikipedia.org/wiki/Deep%20learning?oldid=662755651 Contributors: The Anome, Ed Poor, Michael
Hardy, Meekohi, Glenn, Bearcat, Nandhp, Giraﬀedata, Jonsafari, Oleg Alexandrov, Justin Ormont, BD2412, Qwertyus, Rjwilmsi,
Kri, Bgwhite, Tomdooner, Bhny, Malcolma, Arthur Rubin, Mebden, SeanAhern, Dicklyon, JHP, ChrisCork, Lfstevens, A3nm,
R'n'B, Like.liberation, Popoki, Jshrager, Strife911, Bfx0, Daniel Hershcovich, Pinkpedaller, Dthomsen8, Addbot, Mamikonyana,
Yobot, AnomieBOT, Jonesey95, Zabbarob, Wyverald, RjwilmsiBot, Larry.europe, Helwr, GoingBatty, Sergey WereWolf, SlowByte,
Yoshua.Bengio, JuyangWeng, Renklauf, Bittenus, Widr, BG19bot, Lukas.tencer, Kareltje63, Synchronist, Gameboy97q, IjonTichyIjonTichy, Mogism, AlwaysCoding, Mark viking, Cagarie, Deeper Learning, Prisx, Wikiyant, Underﬂow42, GreyShields, Opokopo, Prof.
Oundest, Gigavanti, Sevensharpnine, Yes deeper, Monkbot, Chieftains337, Samueldg89, Nikunj157, Engheta, Aspurdy, Velvel2, Deng629,
Zhuikov, Stergioc, Jerodlycett, DragonbornXXL and Anonymous: 94
• Feature learning Source: http://en.wikipedia.org/wiki/Feature%20learning?oldid=661746836 Contributors: Phil Boswell, Tobias Bergemann, Qwertyus, Rjwilmsi, Mcld, Kotabatubara, Dsimic, Yobot, AnomieBOT, BG19bot, Mavroudisv, TonyWang0316, Ixjlyons and
Anonymous: 7
• Unsupervised learning Source: http://en.wikipedia.org/wiki/Unsupervised%20learning?oldid=660135356 Contributors: Michael Hardy,
Kku, Alﬁo, Ahoerstemeier, Hike395, Ojigiri~enwiki, Gene s, Urhixidur, Alex Kosorukoﬀ, Aaronbrick, Bobo192, 3mta3, Tablizer, Denoir, Nkour, Qwertyus, Rjwilmsi, Chobot, Roboto de Ajvol, YurikBot, Darker Dreams, Daniel Mietchen, SmackBot, CommodiCast,
Trebor, DHN-bot~enwiki, Lambiam, CRGreathouse, Carstensen, Thijs!bot, Jaxelrod, AnAj, Peteymills, David Eppstein, Agentesegreto, Maheshbest, Timohonkela, Ng.j, EverGreg, Algorithms, Kotsiantis, Auntof6, PixelBot, Edg2103, Addbot, EjsBot, Yobot, Les boys,
AnomieBOT, Salvamoreno, D'ohBot, Skyerise, Ranjan.acharyya, BertSeghers, EmausBot, Fly by Night, Rotcaeroib, Stheodor, Daryakav,
Ida Shaw, Chire, Candace Gillhoolley, WikiMSL, Helpful Pixie Bot, Majidjanz and Anonymous: 40
• Generative model Source: http://en.wikipedia.org/wiki/Generative%20model?oldid=660109222 Contributors: Awaterl, Michael Hardy,
Hike395, Andrewman327, Benwing, Cagri, Rama, Serapio, Jonsafari, Male1979, Qwertyus, Zanetu, Dicklyon, Repied, Barticus88,
RichardSocher~enwiki, Camrn86, YinZhang, Melcombe, Lfriedl, Shanttashjean, Omnipaedista, Geoﬀrey I Webb, Naxingyu, DanielWaterworth, Mekarpeles, ClueBot NG, Gilgoldm, C2oo56, Hestendelin, L T T H U and Anonymous: 14

90

CHAPTER 23. ANDREW NG

• Neural coding Source: http://en.wikipedia.org/wiki/Neural%20coding?oldid=662182422 Contributors: Ed Poor, Apraetor, AndrewKeenanRichardson, LindsayH, CanisRufus, Kghose, Jheald, Woohookitty, BD2412, Rjwilmsi, RDBrown, Colonies Chris, Henrikhenrik, OrphanBot, Radagast83, NickPenguin, Vina-iwbot~enwiki, Clicketyclack, RomanSpa, Sohale, Rji, Andorin, Goodwillein, Anthonyhcole, Nick Number, Davidm617617, Lova Falk, Addbot, Looie496, Luckas-bot, Yobot, Amirobot, Tryptoﬁsh, Citation bot, Xhuo, SassoBot, FrescoBot, Xbcj0843hck3, Albertzeyer, TjeerdB, Jonkerz, Helwr, Bethnim, ZéroBot, Ego White Tray, Bibcode Bot, BG19bot,
FlinZ, ChrisGualtieri, Nicolenewell, Iamozy, Kernsters, Shirindora, Phleg1, Monkbot, PghJesse, Giedroid, AngevineMiller and Anonymous: 33
• Word embedding Source: http://en.wikipedia.org/wiki/Word%20embedding?oldid=623976145 Contributors: Qwertyus, Daniel Hershcovich, Yobot and Citation bot
• Deep belief network Source: http://en.wikipedia.org/wiki/Deep%20belief%20network?oldid=650454380 Contributors: Glenn, Qwertyus, Kri, Like.liberation, Smuckola, Isthmuses, BG19bot, VivamusAmemus, Schurgast and Anonymous: 1
• Convolutional neural network Source: http://en.wikipedia.org/wiki/Convolutional%20neural%20network?oldid=662811606 Contributors: Glenn, Phil Boswell, Bearcat, GregorB, BD2412, Rjwilmsi, Mario23, Serg3d2, Mcld, Chris the speller, Frap, Dr.K., Lfstevens, TheSeven, Like.liberation, Att159, XLinkBot, Yobot, Anypodetos, AnomieBOT, Cugino di mio cugino, Citation bot, Dithridge, Jhbdel, Alvin
Seville, Hobsonlane, RandomDSdevel, Arinelle, Peaceray, Osnetwork, BG19bot, BattyBot, Gameboy97q, APerson, MartinLjungqvist,
Zieglerk, Monkbot, VivamusAmemus, Velvel2, Fight123456, Stri8ted, Gnagyusa, Xksteven and Anonymous: 30
• Restricted Boltzmann machine Source: http://en.wikipedia.org/wiki/Restricted%20Boltzmann%20machine?oldid=650005600 Contributors: Shd~enwiki, Glenn, Dratman, Qwertyus, Kri, Gareth Jones, Deepdraft, Mcld, Hongooi, Dicklyon, Abhineetnazi, Tomtheebomb,
Dsimic, Arsi Warrior, Yobot, LilHelpa, UbaiSandouk, Gilo1969, Racerx11, Leopd, Enerjiparki, Sevensharpnine, Velvel2, Broido,
Dharmablues and Anonymous: 22
• Recurrent neural network Source: http://en.wikipedia.org/wiki/Recurrent%20neural%20network?oldid=662220310 Contributors:
Glenn, Barak~enwiki, Charles Matthews, Urhixidur, Aaronbrick, Tyrell turing, RJFJR, Alai, Male1979, DavidFarmbrough, Seliopou,
Nehalem, Bhny, Rwalker, Banus, That Guy, From That Show!, SmackBot, Moxon, Yume149~enwiki, Charivari, Kissaki0, MichaelGasser,
Terry Bollinger, Dicklyon, Fyedernoggersnodden, Thijs!bot, Daytona2, Curdeius, JaGa, Jamalex~enwiki, KylieTastic, Itb3d, Gdupont,
Jwray, Headlessplatter, Daniel Hershcovich, DumZiBoT, SlaterDeterminant, Achler, Addbot, DOI bot, LatitudeBot, Roux, Fadyone,
Yobot, AnomieBOT, Flewis, Citation bot, Omnipaedista, FrescoBot, Adrian Lange, Citation bot 1, Dmitry St, Epsiloner, Mister Mormon,
DASHBotAV, ClueBot NG, RichardTowers, MerlIwBot, Helpful Pixie Bot, BG19bot, Frze, Justinyap88, Ivan Ukhov, ‫ ,طاها‬Djfrost711,
Randykitty, Jmander, Ukpg, Minky76, OhGodItsSoAmazing, Mknjbhvg, Slashdottir, Monkbot, DoubleDr, Jungkanji and Anonymous: 46
• Long short term memory Source: http://en.wikipedia.org/wiki/Long%20short%20term%20memory?oldid=659668087 Contributors:
Michael Hardy, Glenn, Rich Farmbrough, Denoir, Woohookitty, SmackBot, Derek farn, Ninjakannon, Magioladitis, Barkeep, Pwoolf,
Headlessplatter, M4gnum0n, Muhandes, Jncraton, Yobot, Dithridge, Omnipaedista, Albertzeyer, Silenceisgod, Epsiloner, Ego White Tray,
Mister Mormon, Hmainsbot1, Mogism, Velvel2 and Anonymous: 13
• Google Brain Source: http://en.wikipedia.org/wiki/Google%20Brain?oldid=657438840 Contributors: Bearcat, Dfrankow, Vipul, Daranz,
JorisvS, Dicklyon, Aeternus, Diaa abdelmoneim, AnomieBOT, Cnwilliams, Chire, BG19bot, Q6637p, Stephen Balaban, BattyBot, Blharp and Anonymous: 8
• Google DeepMind Source: http://en.wikipedia.org/wiki/Google%20DeepMind?oldid=654694254 Contributors: Ciphergoth, Bearcat,
Dratman, Bgwhite, Roboﬁsh, Aeternus, Edwardx, Yellowdesk, Magioladitis, Touch Of Light, Steel1943, Ciphershort, Wikiisawesome,
Sprachpﬂeger, Green Cardamom, Larry.europe, BG19bot, IjonTichyIjonTichy, Megab, William 2239, Bipper1024, Jon Jonathan, Harrisonpop and Anonymous: 13
• Torch (machine learning) Source: http://en.wikipedia.org/wiki/Torch%20(machine%20learning)?oldid=661012919 Contributors: Qwertyus, Strife911, Niceguyedc, Larry.europe, BG19bot, Lor and Anonymous: 6
• Theano (software) Source: http://en.wikipedia.org/wiki/Theano%20(software)?oldid=641284109 Contributors: Bearcat, Qwertyus, Andre.holzner, Strife911, Mrocklin, MartinThoma, Turn685 and Monkbot
• Deeplearning4j Source: http://en.wikipedia.org/wiki/Deeplearning4j?oldid=662808219 Contributors: Rwalker, Cydebot, Like.liberation,
NinjaRobotPirate, Daniel Hershcovich, Dawynn, Yobot, Wcherowi and Anonymous: 3
• Gensim Source: http://en.wikipedia.org/wiki/Gensim?oldid=651026231 Contributors: Thorwald, Qwertyus, Dawynn, Matěj Grabovský,
Smk65536, Armbrust, Lightlowemon, Delusion23, Velvel2 and Anonymous: 10
• Geoﬀrey Hinton Source: http://en.wikipedia.org/wiki/Geoffrey%20Hinton?oldid=660779435 Contributors: Edward, Zeno Gantner,
Rainer Wasserfuhr~enwiki, Flockmeal, MOiRe, Diberri, Risk one, Just Another Dan, Lawrennd, Rich Farmbrough, Davidswelt, Marudubshinki, Qwertyus, Rjwilmsi, Winterstein, Misterwindupbird, RussBot, Welsh, Gareth Jones, DaveWF, BorgQueen, InverseHypercube,
CRKingston, Jsusskin, Onkelschark, OrphanBot, Jmlk17, Dl2000, Shoeofdeath, Aeternus, Adam Newbold, Cydebot, Michael Fourman,
Roweis, Waacstats, Destynova, MetsBot, David Eppstein, STBot, FMAFan1990, AlexGreat, Sidcool1234, XLinkBot, Galyet, G7valera,
Addbot, ‫ ,דוד שי‬Omnipaedista, Plucas58, Morton Shumway, RjwilmsiBot, Larry.europe, Gumbys, Wpeaceout, Onionesque, BG19bot,
ChrisGualtieri, Makecat-bot, Anne Delong, Silas Ropac, Putting things straight, Annaﬂagg, Justanother109, Jonarnold1985, Velvel2, Mathewk1300 and Anonymous: 24
• Yann LeCun Source: http://en.wikipedia.org/wiki/Yann%20LeCun?oldid=662361291 Contributors: Zeno Gantner, Klemen Kocjancic,
Rich Farmbrough, Runner1928, Rschen7754, Bgwhite, RussBot, Gareth Jones, Jpbowen, Angoodkind, Cydebot, Studerby, Bongwarrior,
Waacstats, 72Dino, Aboutmovies, Falcon8765, Profshadoko, Cajunbill, M4gnum0n, Berean Hunter, MystBot, Addbot, Yobot, Bunnyhop11, FrescoBot, Lbottou, Jvsarun1993, Uprightmanpower, JJRambo, Zzym, Hgsa5, AdjSilly, Polochicken, Lljjp, Crickavery, Justanother109, Voltdye, Velvel2, Visionscholar, Mathewk1300, KasparBot, Algorith and Anonymous: 3
• Jürgen Schmidhuber Source: http://en.wikipedia.org/wiki/J%C3%BCrgen%20Schmidhuber?oldid=660776145 Contributors: Michael
Hardy, Kosebamse, Hike395, Charles Matthews, Wik, Randomness~enwiki, Goedelman, Psb777, Newbie~enwiki, Juxi, Robin klein,
Klemen Kocjancic, D6, On you again, Kelly Ramsey, Rpresser, Derumi, Woohookitty, GregorB, Male1979, FlaBot, Bgwhite, RussBot,
SmackBot, Hongooi, Ben Moore, Phoxhat, JimStyle61093475, IDSIAupdate, Mpotse, Blaisorblade, Marek69, A1bb22, Waacstats, JaGa,
Gwern, Kornfan71, R'n'B, DadaNeem, BOTijo, Noveltyghost, Tbsdy lives, Addbot, Lightbot, Yobot, Fleabox, Omnipaedista, Underlying lk, RjwilmsiBot, Angrytoast, Epsiloner, Helpful Pixie Bot, VIAFbot, It's here!, Midnightplunge, Laanaae, Hildensia, KasparBot and
Anonymous: 16

23.7. TEXT AND IMAGE SOURCES, CONTRIBUTORS, AND LICENSES

91

• Jeﬀ Dean (computer scientist) Source: http://en.wikipedia.org/wiki/Jeff%20Dean%20(computer%20scientist)?oldid=657563156 Contributors: Klemen Kocjancic, Lockley, AntonioDsouza, Wavelength, Gareth Jones, Elkman, CommonsDelinker, Aboutmovies, Mr.Z-bot,
Omnipaedista, Patchy1, Iamjamesbond007, BG19bot, Stephen Balaban, Stephenbalaban, Timothy Gu, ChrisGualtieri, Mogism and Anonymous: 9
• Andrew Ng Source: http://en.wikipedia.org/wiki/Andrew%20Ng?oldid=659094106 Contributors: Michael Hardy, Msm, Simon LacosteJulien, Prenju, Klemen Kocjancic, Rich Farmbrough, Vipul, Rd232, Mr Tan, Mandarax, Alex Bakharev, Gareth Jones, Pyronite, Johndburger, InverseHypercube, Smallbones, Raysonho, CmdrDan, Cydebot, Batra, Utopiah, Magioladitis, Rootxploit, CommonsDelinker,
Coolg49964, Bcnof, Hoising, Maghnus, XKL, Arbor to SJ, Martarius, Chaosdruid, XLinkBot, Addbot, Mortense, Yobot, Azylber,
AnomieBOT, Chrisvanlang, Omnipaedista, Velblod, FrescoBot, Abductive, Amhey, RjwilmsiBot, Mmm333k, ZéroBot, Bemanna, ClueBot NG, Kashthealien, Geistcj, Rrrlf, ArmbrustBot, Happyhappy001, E8xE8, Gnomy7, Linuxjava, AcrO O, Calisacole, Velvel2, Visionscholar, Demdim0, Csisawesome and Anonymous: 41

23.7.2

Images

• File:Ambox_important.svg Source: http://upload.wikimedia.org/wikipedia/commons/b/b4/Ambox_important.svg License: Public domain Contributors: Own work, based oﬀ of Image:Ambox scales.svg Original artist: Dsmurat (talk · contribs)
• File:Ann_dependency_(graph).svg Source: http://upload.wikimedia.org/wikipedia/commons/d/dd/Ann_dependency_%28graph%29. svg License: CC BY-SA 3.0 Contributors: Vector version of File:Ann dependency graph.png Original artist: Glosser.ca
• File:Colored_neural_network.svg Source: http://upload.wikimedia.org/wikipedia/commons/4/46/Colored_neural_network.svg License: CC BY-SA 3.0 Contributors: Own work, Derivative of File:Artificial neural network.svg Original artist: Glosser.ca
• File:Commons-logo.svg Source: http://upload.wikimedia.org/wikipedia/en/4/4a/Commons-logo.svg License: ? Contributors: ? Original artist: ?
• File:Deep_belief_net.svg Source: http://upload.wikimedia.org/wikipedia/commons/f/fa/Deep_belief_net.svg License: CC0 Contributors:
Own work Original artist: Qwertyus
• File:Edit-clear.svg Source: http://upload.wikimedia.org/wikipedia/en/f/f2/Edit-clear.svg License: Public domain Contributors: The
Tango! Desktop Project. Original artist:
The people from the Tango! project. And according to the meta-data in the ﬁle, speciﬁcally:“Andreas Nilsson, and Jakub Steiner (although minimally).” • File:Elman_srnn.png Source: http://upload.wikimedia.org/wikipedia/commons/8/8f/Elman_srnn.png License: CC BY 3.0 Contributors:
Own work Original artist: Fyedernoggersnodden
• File:Emoji_u1f4bb.svg Source: http://upload.wikimedia.org/wikipedia/commons/d/d7/Emoji_u1f4bb.svg License: Apache License 2.0
Contributors: https://code.google.com/p/noto/ Original artist: Google
• File:Fisher_iris_versicolor_sepalwidth.svg Source: http://upload.wikimedia.org/wikipedia/commons/4/40/Fisher_iris_versicolor_ sepalwidth.svg License: CC BY-SA 3.0 Contributors: en:Image:Fisher iris versicolor sepalwidth.png Original artist: en:User:Qwfp (original); Pbroks13 (talk) (redraw)
• File:Folder_Hexagonal_Icon.svg Source: http://upload.wikimedia.org/wikipedia/en/4/48/Folder_Hexagonal_Icon.svg License: Cc-bysa-3.0 Contributors: ? Original artist: ?
• File:Free_Software_Portal_Logo.svg Source: http://upload.wikimedia.org/wikipedia/commons/3/31/Free_and_open-source_ software_logo_%282009%29.svg License: Public domain Contributors: FOSS Logo.svg Original artist: Free Software Portal Logo.svg
(FOSS Logo.svg): ViperSnake151
• File:Gensim_logo.png Source: http://upload.wikimedia.org/wikipedia/en/b/b1/Gensim_logo.png License: Fair use Contributors: http://radimrehurek.com/gensim/_static/images/logo-gensim.png Original artist: ?
• File:Internet_map_1024.jpg Source: http://upload.wikimedia.org/wikipedia/commons/d/d2/Internet_map_1024.jpg License: CC BY
2.5 Contributors: Originally from the English Wikipedia; description page is/was here. Original artist: The Opte Project
• File:Lstm_block.svg Source: http://upload.wikimedia.org/wikipedia/en/8/8d/Lstm_block.svg License: PD Contributors:
Headlessplatter (talk) (Uploads) Original artist:
Headlessplatter (talk) (Uploads)
• File:NoisyNeuralResponse.png Source: http://upload.wikimedia.org/wikipedia/en/6/66/NoisyNeuralResponse.png License: PD Contributors: ? Original artist: ?
• File:Nuvola_apps_arts.svg Source: http://upload.wikimedia.org/wikipedia/commons/e/e2/Nuvola_apps_arts.svg License: GFDL Contributors: Image:Nuvola apps arts.png Original artist: Manco Capac
• File:P_vip.svg Source: http://upload.wikimedia.org/wikipedia/en/6/69/P_vip.svg License: PD Contributors: ? Original artist: ?
• File:People_icon.svg Source: http://upload.wikimedia.org/wikipedia/commons/3/37/People_icon.svg License: CC0 Contributors: OpenClipart Original artist: OpenClipart
• File:PopulationCode.svg Source: http://upload.wikimedia.org/wikipedia/commons/a/a1/PopulationCode.svg License: Public domain
Contributors: Image:PopulationCode.png at English Wikipedia Original artist:
• Original: AndrewKeenanRichardson
• File:Portal-puzzle.svg Source: http://upload.wikimedia.org/wikipedia/en/f/fd/Portal-puzzle.svg License: Public domain Contributors: ?
Original artist: ?
• File:Question_book-new.svg Source: http://upload.wikimedia.org/wikipedia/en/9/99/Question_book-new.svg License: Cc-by-sa-3.0
Contributors:
Created from scratch in Adobe Illustrator. Based on Image:Question book.png created by User:Equazcion Original artist:
Tkgd2007

92

CHAPTER 23. ANDREW NG

• File:Recurrent_ann_dependency_graph.png Source: http://upload.wikimedia.org/wikipedia/commons/7/79/Recurrent_ann_ dependency_graph.png License: CC-BY-SA-3.0 Contributors: ? Original artist: ?
• File:Restricted_Boltzmann_machine.svg Source: http://upload.wikimedia.org/wikipedia/commons/e/e8/Restricted_Boltzmann_ machine.svg License: CC BY-SA 3.0 Contributors: Own work Original artist: Qwertyus
• File:Science-symbol-2.svg Source: http://upload.wikimedia.org/wikipedia/commons/7/75/Science-symbol-2.svg License: CC BY 3.0
Contributors: en:Image:Science-symbol2.png Original artist: en:User:AllyUnion, User:Stannered
• File:Synapse_deployment.jpg Source: http://upload.wikimedia.org/wikipedia/en/2/22/Synapse_deployment.jpg License: CC-BY-SA2.5 Contributors: ? Original artist: ?
• File:Text_document_with_red_question_mark.svg Source: http://upload.wikimedia.org/wikipedia/commons/a/a4/Text_document_ with_red_question_mark.svg License: Public domain Contributors: Created by bdesham with Inkscape; based upon Text-x-generic.svg from the Tango project. Original artist: Benjamin D. Esham (bdesham)
• File:Torch_2014_logo.png Source: http://upload.wikimedia.org/wikipedia/en/f/f5/Torch_2014_logo.png License: Fair use Contributors: https://github.com/torch Original artist: ?
• File:Wave.svg Source: http://upload.wikimedia.org/wikipedia/commons/4/40/Wave.svg License: BSD Contributors: http://duke.kenai. com/wave/index.html (new), https://duke.dev.java.net/images/wave/index.html (old) Original artist: sbmehta converted to SVG from Sun
Microsystems AI version.
• File:Wikibooks-logo-en-noslogan.svg Source: http://upload.wikimedia.org/wikipedia/commons/d/df/Wikibooks-logo-en-noslogan. svg License: CC BY-SA 3.0 Contributors: Own work Original artist: User:Bastique, User:Ramac et al.
• File:Wikiversity-logo.svg Source: http://upload.wikimedia.org/wikipedia/commons/9/91/Wikiversity-logo.svg License: CC BY-SA 3.0
Contributors: Snorky (optimized and cleaned up by verdy_p) Original artist: Snorky (optimized and cleaned up by verdy_p)

23.7.3

Content license

• Creative Commons Attribution-Share Alike 3.0

Deep Learning Wikipedia

Similar Documents

It Is Achievable

Integration of Technology

Bp Case

Ishikawa Analysis

Informative Speech On Mental Golf

Start a New Career

Writing for Fun

Eng 125

Artificial Intelligence

How Something Can Come from “Nothing”

Rooom Rooom

Ben Vereen

Requirements for Drm Research Paper

Psy-101

Angles in America

Popular Essays