Figures From �Artificial Intelligence Engines

Some figures are copied with permission from other sources (see book).

​

Other figures created by JV Stone for this book are are licensed under a Creative Commons Attribution-Non Commercial 4.0 International License.

JV Stone

j.v.stone@sheffield.ac.uk

Figure 1.2

Figure 1.1

Figure 1.3

Figure 1.4

Perceptron

Backprop network

Hopfield Net

Boltzmann machine

Deep neural network

Figure 1.5

Figure 1.6

Figure 1.7

Figure 2.1

Figure 2.2

Figure 2.3

Figure 2.4

ΔE

Δw

gradient = ΔE/Δw

w*

w

Figure 2.5

Figure 2.6

Figure 2.7

Figure 2.8

Figure 3.1

Figure 3.2

Figure 3.3

Figure 3.4

Figure 3.5

.

θ

Classified as C₁

Classified as C₂

x

w.x

Decision

boundary

.

.

.

.

.

.

.

.

.

.

.

x₁

x₂

α

.

.

.

.

w

Figure 3.6

-1 +1

+1

-1

a

Weight

Vector

x

w

x1

x2

=

=

=

=

=

=

x

Figure 3.7

Figure 4.1

Figure 4.2

Figure 4.3

Figure 4.4

Figure 4.5

Figure 4.6

Figure 4.7

Figure 4.8

Figure 4.9

Figure 4.10

w₁

w₂

Local

minimum

Global

minimum

E(w)

Figure 4.11

Error function E

Weight

Global

minimum

Local

minima

Figure 4.12

-- i -- c o u l

Phoneme = /k/

Input units

Hidden units

Output units

Figure 4.13

Figure 4.14

Figure 5.1

Figure 5.2

Figure 6.1

Outer loop

​

Figure 6.2

Figure 6.3

Figure 6.4

Figure 6.5

Figure 7.1

Figure 7.2

Figure 7.3

Figure 7.4

Figure 7.5

Figure 7.6

Figure 7.7

Figure 7.8

Figure 8.1

Encoder

Encoder

network

input = x

Latent variables

Figure 8.2

Figure 8.3

z₁

.

encode

decode

z₂

input = x

output = x’

Latent space

Figure 8.4

Figure 8.5

z₁

z₂

Figure 8.6

Figure 8.7

Encoder

Encoder

network

Decoder

network

Decoder

z_i

μ_t

Σ_t

Sampled vector

z_i ~ q(z|x_t)

x_t ~ p(x)

input = x_t

output = x_t’

q(z|x_t)

Figure 8.8

Decoder

x₁

x₂

z₁

z₂

μ₁

.

μ₂

.

q(z|x₁)

q(z|x₂)

p(x’|μ₁)

p(x’|μ₂)

Figure 8.9

w*

L

Weight

Log probability

log p(x)

D(q(z|x)||p(z|x))

Figure 8.10

x’

p(x|x’)

p(x’)

x

Probability

Figure 9.1

Ganglion cell

outputs

Cone outputs

Ganglion cell

receptive field

Ganglion cell

Cone

Retinal image

Figure 9.2

Hidden layer

Pixels

(visible layer)

Receptive field

Visible unit output

​

Hidden unit output

​

Visible unit input

Image

Hidden layer outputs

Connection weight

Hidden unit input

​

Figure 9.3

Figure 9.4

Figure 9.5

Input

Layer

16x16 image

Hidden

Layer 1

12 8x8 feature maps

Hidden

Layer 2

12 4x4

feature

maps

Hidden

Layer 3

30 units

5x5

convolution

5x5

convolution

​

Output

Layer

10 units

Figure 9.6

Figure 9.7

Figure 9.8

Figure 9.8

+

x

H(x) = F(x) + x

H(x) = F(x) + x

F(x) = H(x) - x

Input layer

Hidden layer

Figure 9.10

y

ŷ

noise

Noisy Encoder

Decoder

Encoder

Figure 9.11

?

b

c

d

Class A

Class B

Class A

Class A

Class A

Class B

Class B

Class B

X

Figure 9.12

Figure 9.13

Discriminator

Generator

Random

noise z

Training

set

Generated

image

Decision

image x

Figure 9.13

Figure 9.14

Figure 10.1

Figure 10.2

Figure 10.3

Figure 10.4

v* π*

Improving V

Improving π

Estimate V with fixed π

Estimate π with fixed V

Optimal v and π

Initial estimate

of policy and

state-value function

Estimate π with fixed V

Figure 10.5

Environment

Policy

State-Value

Function

Reward

Actor

Action

TD error

State

Critic

Figure 10.6

Figure 10.7

The end