CBP 2009-10Comp 3104 The Nature of Computing 1 Nature Inspired Computing Artificial Neural Nets -...

20
CBP 2009-10 Comp 3104 The Nature of C omputing 1 Nature Inspired Computing ficial Neural Nets - Symbiosis between computer and cognitive scienc

Transcript of CBP 2009-10Comp 3104 The Nature of Computing 1 Nature Inspired Computing Artificial Neural Nets -...

Page 1: CBP 2009-10Comp 3104 The Nature of Computing 1 Nature Inspired Computing Artificial Neural Nets - Symbiosis between computer and cognitive sciences.

CBP 2009-10 Comp 3104 The Nature of Computing

1

Nature Inspired ComputingArtificial Neural Nets - Symbiosis between computer and cognitive sciences

Page 2: CBP 2009-10Comp 3104 The Nature of Computing 1 Nature Inspired Computing Artificial Neural Nets - Symbiosis between computer and cognitive sciences.

CBP 2009-10 Comp 3104 The Nature of Computing

2

Cajal -1-

Cajal Golgi

Cajal + Golgi indentification of independent neurons by staining, microscopy and looking. (Nobel Prize 1906)

Page 3: CBP 2009-10Comp 3104 The Nature of Computing 1 Nature Inspired Computing Artificial Neural Nets - Symbiosis between computer and cognitive sciences.

CBP 2009-10 Comp 3104 The Nature of Computing

3

Cajal -2-

Page 4: CBP 2009-10Comp 3104 The Nature of Computing 1 Nature Inspired Computing Artificial Neural Nets - Symbiosis between computer and cognitive sciences.

CBP 2009-10 Comp 3104 The Nature of Computing

4

Rat Neurons

Page 5: CBP 2009-10Comp 3104 The Nature of Computing 1 Nature Inspired Computing Artificial Neural Nets - Symbiosis between computer and cognitive sciences.

CBP 2009-10 Comp 3104 The Nature of Computing

5

Neuron Grows on Electrodes

Copyright © 2000 Yoonkey Nam - Department of Electrical & Computer Engineering; Imaging Technology Group; Beckman Institute; and the University of Illinois

Page 6: CBP 2009-10Comp 3104 The Nature of Computing 1 Nature Inspired Computing Artificial Neural Nets - Symbiosis between computer and cognitive sciences.

CBP 2009-10 Comp 3104 The Nature of Computing

6

Investigation of Single Neurons

Microelectrode recording of Biological Neuron activation using tungsten electrode

Hubel and Weisel. Nobel Prize 1958

Photomicrograph: Height = 1mm.

Page 7: CBP 2009-10Comp 3104 The Nature of Computing 1 Nature Inspired Computing Artificial Neural Nets - Symbiosis between computer and cognitive sciences.

CBP 2009-10 Comp 3104 The Nature of Computing

7

Biological Neurons

dendrites

axonsynapse

Signal flow Big Neurological principle #1 Neurons work using electricity, not blood or other special goo

Signal shape

Page 8: CBP 2009-10Comp 3104 The Nature of Computing 1 Nature Inspired Computing Artificial Neural Nets - Symbiosis between computer and cognitive sciences.

CBP 2009-10 Comp 3104 The Nature of Computing

8

Single Neuron

In 2

In 1

In 3

In 4In 1 + In 2 + In 3 + In 4

“input”

“activation”

“activation”

“input”

A

A

B

B

“threshold”

Big Neurological principle #2 “Integrate and Fire” Inputs summed. If above threshold output fires.

Page 9: CBP 2009-10Comp 3104 The Nature of Computing 1 Nature Inspired Computing Artificial Neural Nets - Symbiosis between computer and cognitive sciences.

CBP 2009-10 Comp 3104 The Nature of Computing

9

Learning in Neural Nets

Before Learning

After Learning

Big Neurological principle #3 “Hebbian Learning” Synapse strength increases if both cells A and B are firing

A

A

B

B

Page 10: CBP 2009-10Comp 3104 The Nature of Computing 1 Nature Inspired Computing Artificial Neural Nets - Symbiosis between computer and cognitive sciences.

CBP 2009-10 Comp 3104 The Nature of Computing

10

Brains Minds and Computers

Brains Computers

• Work using Electricity• Have inputs and outputs

• Can learn by experience• Can be taught

• Work using Electricity• Have inputs and outputs

• Can be programmed

So do we understand brains? Yep. Do we therefore understand Minds? Nope.

Page 11: CBP 2009-10Comp 3104 The Nature of Computing 1 Nature Inspired Computing Artificial Neural Nets - Symbiosis between computer and cognitive sciences.

CBP 2009-10 Comp 3104 The Nature of Computing

11

Artificial Neurons

In 1 + In 2 + In 3 + In 4

In 1

In 2

In 3

“output”

“input”

A

A

B

B

“threshold”

inputs

output

Page 12: CBP 2009-10Comp 3104 The Nature of Computing 1 Nature Inspired Computing Artificial Neural Nets - Symbiosis between computer and cognitive sciences.

CBP 2009-10 Comp 3104 The Nature of Computing

12

Learning Logical Gates

0.5

Threshold = 1

1

Threshold = 1

1

1.5

Ouput neuron fires only when sum is greater than the threshold

AND - gate

OR - gate

A B O

0 0 0

0 1 0

1 0 0

1 1 1

A B O

0 0 0

0 1 1

1 0 1

1 1 1

Page 13: CBP 2009-10Comp 3104 The Nature of Computing 1 Nature Inspired Computing Artificial Neural Nets - Symbiosis between computer and cognitive sciences.

CBP 2009-10 Comp 3104 The Nature of Computing

13

Training an Artificial Neural Net

eyes motors

right

left

1. We want to train the robot to move towards the light

2. So when the right eye gets light, the left motor neuron must fire and vice versa

3. We must therefore strengthen the cross - connections and kill the direct connections

Page 14: CBP 2009-10Comp 3104 The Nature of Computing 1 Nature Inspired Computing Artificial Neural Nets - Symbiosis between computer and cognitive sciences.

CBP 2009-10 Comp 3104 The Nature of Computing

14

Back Propagation of Errors

eyes motors

right

left

1 0.5

eyes motors

right

left

1

0.5

Let’s say desired motor drive is 1.0 (full forwards).

Here the right motor drive should be 0 but it is 0.5. So the error is

Desired - actual = 0.0 - 0.5 = -0.5

So we decrease the connection by 0.5

Here the right motor drive should be 1.0 but it is 0.5. So the error is

Desired - actual = 1.0 - 0.5 = +0.5

So we increase the connection by 0.5

Page 15: CBP 2009-10Comp 3104 The Nature of Computing 1 Nature Inspired Computing Artificial Neural Nets - Symbiosis between computer and cognitive sciences.

CBP 2009-10 Comp 3104 The Nature of Computing

15

Neural Net Solver

Page 16: CBP 2009-10Comp 3104 The Nature of Computing 1 Nature Inspired Computing Artificial Neural Nets - Symbiosis between computer and cognitive sciences.

CBP 2009-10 Comp 3104 The Nature of Computing

16

Medical Application

Flu

Neural Net

cough

headache

Page 17: CBP 2009-10Comp 3104 The Nature of Computing 1 Nature Inspired Computing Artificial Neural Nets - Symbiosis between computer and cognitive sciences.

CBP 2009-10 Comp 3104 The Nature of Computing

17

Medical Diagnosis

Cough

Headache

Meningitis

Flu

Pneuomonia

Not ill

1Cough

Headache 1

Flu 1

Page 18: CBP 2009-10Comp 3104 The Nature of Computing 1 Nature Inspired Computing Artificial Neural Nets - Symbiosis between computer and cognitive sciences.

CBP 2009-10 Comp 3104 The Nature of Computing

18

“Classical” Medical Diagnosis

If ( (symptom ! = cough) && (symptom != headache) )

illness = no illness;

else if ( (symptom ! = cough) && (symptom == headache) )

illness = meningitis;

else if ( (symptom == cough) && (symptom != headache) )

illness = pneumonia;

else if ( (symptom == cough) && (symptom == headache) )

illness = flu;

Rule-based Learning “ if … then …. else … “

Page 19: CBP 2009-10Comp 3104 The Nature of Computing 1 Nature Inspired Computing Artificial Neural Nets - Symbiosis between computer and cognitive sciences.

CBP 2009-10 Comp 3104 The Nature of Computing

19

ECG Interpretation

R-R interval

S-T elevation

P-R interval

QRS duration

AVF lead

QRS amplitude

SV tachycardia

Ventricular tachycardia

LV hypertrophy

RV hypertrophy

Myocardial infarction

Page 20: CBP 2009-10Comp 3104 The Nature of Computing 1 Nature Inspired Computing Artificial Neural Nets - Symbiosis between computer and cognitive sciences.

CBP 2009-10 Comp 3104 The Nature of Computing

20

NNets vs Expert Systems

Modeling ExamplesExplanation

Effort Needed Provided

Rule-based Exp. Syst. high low high

Bayesian Nets high low moderate

Classification Trees low high “high”

Neural Nets low high low

Regression Models high moderate moderate