Artificial Intelligence and Learning Algorithms: Presented by Brian M. Frezza 12/1/05

Artificial Intelligence and Learning Algorithms
Presented By Brian M. Frezza 12/1/05
Game Plan
Whats a Learning Algorithm? Why should I care?
Biological parallels
Real World Examples Getting our hands dirty with the algorithms
Bayesian Networks Hidden Markov Models Genetic Algorithms Neural Networks
Artificial Neural Networks Vs Neuron Biology

Frasers Rules
Frontiers in AI
Hard Math
Whats a Learning Algorithm?

An algorithm which predicts datas future behavior based on its past performance.
Programmer can be ignorant of the datas trends.
Not rationally designed!
Training Data Test Data
Why do I care?
Use In Informatics
Predict trends in fuzzy data
Subtle patterns in data Complex patterns in data Noisy data
Network inference Classification inference
Analogies To Chemical Biology

Evolution Immunological Response Neurology
Fundamental Theories of Intelligence

Thats heavy dude
Street Smarts
CMUs Navlab-5 (No Hands Across America)
1995 Neural Network Driven Car Pittsburgh to San Diego: 2,797 miles (98.2%) Single hidden layer backpropagation network!
Subcellular location through fluorescence

A Neural network classifier capable of recognizing the patterns of all major subcellular structures in fluorescence microscope images of HeLa cells M. V. Boland, and R. F. Murphy, Bioinformatics (2001) 17(12), 1213-1223
Protein secondary structure prediction Intron/Exon predictions Protein/Gene network inference Speech recognition Face recognition
The Algorithms
Bayesian Networks: Basics

Requires models of how data behaves
Set of Hypothesis: {H}
Keeps track of likelihood of each model being accurate as data becomes available
P(H)
Predicts as a weighted average

P(E) = Sum( P(H)*H(E) )
Bayesian Network Example

What color hair will Paul Schaffers kids have if he marries Redhead?
Hypothesis
Ha(rr) rr x rr: 100% Redhead Hb(Rr) rr x Rr: 50% Redhead 50% Not Hc(RR) rr x RR: 100% Not
Initially clueless:
So P(Ha) = P(Hb) = P(Hc) = 1/3
Bayesian Network: Trace

History Hypothesis Ha: 100% Redhead Hb: 50% Redhead 50% Not Hc: 100% Not
Redhead 0 Not 0
P(Ha)
Likelihood's
P(Hb)
P(Hc)
1/3
1/3
1/3
Prediction: Will their next kid be a Redhead?
= P(red|Ha)*P(Ha) + P(red|Hb)*P(Hb) + P(red|Hc)*P(Hc) = (1)*(1/3) + (1/2)*(1/3) + (0)(1/3) =(1/2)
Bayesian Network:Trace
Redhead 1 Not 0
P(Ha)
Likelihood's
P(Hb)
P(Hc)
1/2
1/2

Redhead 2 Not 0
P(Ha)
Likelihood's
P(Hb)
P(Hc)
3/4
1/4

Redhead 3 Not 0
P(Ha)
Likelihood's
P(Hb)
P(Hc)
7/8
1/8
Bayesian Networks Notes

Never reject hypothesis unless directly disproved Learns based on rational models of behavior
Models can be extracted!
Programmer needs to form hypothesis beforehand.
The Algorithms
Hidden Markov Models(HMM)

Discrete learning algorithm
Programmer must be able to categorize predictions
HMMs also assume a model of the world working behind the data Models are also extractable Common Uses
Speech Recognition Secondary structure prediction Intron/Exon predictions Categorization of data
Hidden Markov Models: Take a Step Back

1st order Markov Models:
Q{States} Pr{Transition} Sum of all P(T) out of state = 1
P3 1 P2
P1
Q2
Q1
Q3
1-P3
1-P1-P2
P4
Q4
1-P4
1st order Markov Model Setup

Pick Initial state: Q1 Pick Transition Probabilities:
P1 P2 P3 P4
0.6
0.2
0.9
0.4
Q2
1 P2 P3
For each time step

Pick a random number 0.0-1.0
Q1
P1
Q3
1-P3
1-P1-P2
P4
Q4
1-P4
1st order Markov Model Trace

Current State: Q1 Transition probabilities:
P1 0.6 P2 0.2 P3 0.9 P4 0.4
P1 P3
Time Step = 1
Random Number:
0.22341
Q2
1
P2 1-P1-P2 1-P3
So Next State:
0.22341 < P1
Q2
Take P1
P4
Q1
Q3
Q4
1-P4

P1 0.6 P2 0.2 P3 0.9 P4 0.4
P1 P3
Time Step = 2
Random Number:
0.64357
Q2
1
P2 1-P1-P2 1-P3
So Next State:
No Choice, P = 1 Q3
P4
Q1
Q3
Q4
1-P4

P1 0.6 P2 0.2 P3 0.9 P4 0.4
P1 P3
Time Step = 3
Random Number:
0.97412
Q2
1
P2 1-P1-P2 1-P3
So Next State:
0.97412 > 0.9
Q4
Take 1-P3
P4
Q1
Q3
Q4
1-P4

P1 P2 P3 P4
Time Step = 4
0.6
0.2
0.9
0.4
P1
Im going to stop here. Markov Chain:

Q1, Q2, Q3, Q4
P4
Q2
1
P2 1-P1-P2 1-P3
P3
Q1
Q3
Q4
1-P4
What else can Markov do?

Higher Order Models
Kth order
Metropolis-Hastings
Determining thermodynamic equilibrium
Continuous Markov Models

Time step varies according to continuous distribution
Hidden Markov Models

Discrete model learning
Hidden Markov Models (HMMs)

A Markov Model drives the world but it is hidden from direct observation and its status must be inferred from a set of observables.
Voice recognition
Observable: Sound waves Hidden states: Words
Intron/Exon prediction
Observable: nucleotide sequence Hidden State: Exon, Intron, Non-coding
Secondary structure prediction for protein

Observable: Amino acid sequence Hidden State: Alpha helix, Beta Sheet, Unstructured
Hidden Markov Models: Example

Secondary Structure Prediction
His Asp Arg Phe Ser Ala Tyr Cis Thr Ser Ile Gln Trp Glu Pro Lys Gly Leu Met Asn Val
Observable States
Hidden States
Alpha Helix
Unstructured
Beta Sheet
Hidden Markov Models: Smaller Example Exon/Intron Mapping

A
P(A|Ex) P(T|Ig) P(A|Ig) P(G|Ig) P(C|Ig) P(A|It) P(T|Ex)
P(T|It)
P(G|Ex)
P(G|It)
P(C|Ex)
C
P(C|It)
Observable States
Hidden States
P(Ex|Ex) P(In|Ex) P(Ig|It) P(It|It)
Intergenic
Exon
P(Ex|Ig)
P(Itr|Ig)
Intron
P(Ex|It)
P(In|Ex)
P(Ig|Ig)
Hidden Markov Models: Smaller Example Exon/Intron Mapping

Hidden State Transition Probabilities Observable State Probabilities
To Hidden State
Observable
Ex Ig It
Ex Ig It 0.7 0.1 0.2 0.49 0.5 0.01 0.18 0.02 0.8
From
Ex 0.33 0.42 0.11 0.14 Ig 0.25 0.25 0.25 0.25 It

0.14 0.16 0.5 0.2
Starting Distribution
Ex 0.1
Ig 0.89
It 0.01
Hidden Markov Model

How to predict outcomes from a HMM Brute force: Try every possible Markov chain
Which chain has greatest probability of generating observed data?
Viterbi algorithm
Dynamic programming approach
Viterbi Algorithm: Trace

Hidden State Transition Probabilities To
Ex Ig 0.1 0.9 0.02 It 0.2 0.01 0.8 A T A Ig It 0.09 0.18 A T G
Example Sequence: ATAATGGCGAGTG

Exon 3.3*10-2 Introgenic 2.2*10-1 Intron 1.4*10-3
From
Ex
0.7
Observable State Probabilities Observable A T

0.42 0.25 0.16
G C G
G
0.11 0.25 0.5
C
0.14 0.25 0.2
A G A T
Hidden State
Ex Ig It
0.33 0.25 0.14
Ex 0.1 Ig 0.89 It 0.01
Exon = P(A|Ex) * Start Exon = 3.3*10-2 Introgenic = P(A|Ig) * Start Ig = 2.2*10-1 Intron = P(A|It) * Start It = 0.14 * 0.01 = 1.4*10-3

A Ex Ig 0.1 0.5 0.02 It 0.2 0.01 0.8 T A Ex Ig It 0.7 0.49 0.18

Exon 3.3*10-2 4.6*10-2 Introgenic 2.2*10-1 2.8*10-2 Intron 1.4*10-3 1.1*10-3
From
A
T G

0.42 0.25 0.16
G C G
G
0.11 0.25 0.5
C
0.14 0.25 0.2
A G A
Hidden State
Ex Ig It
0.33 0.25 0.14
T G Exon = Max( P(Ex|Ex)*Pn-1(Ex), P(Ex|Ig)*Pn-1(Ig), P(Ex|It)*Pn-1(It) ) *P(T|Ex) = 4.6*10-2 Introgenic =Max( P(Ig|Ex)*Pn-1(Ex), P(Ig|Ig)*Pn-1(Ig), P(Ig|It)*Pn-1(It) ) * P(T|Ig) = 2.8*10-2 Intron = Max( P(It|Ex)*Pn-1(Ex), P(It|Ig)*Pn-1(Ig), P(It,It)*Pn-1(It) ) * P(T|It) = 1.1*10-3
Ex 0.1 Ig 0.89 It 0.01


Exon 3.3*10-2 4.6*10-2 1.1*10-2 Introgenic 2.2*10-1 2.8*10-2 3.5*10-3 Intron 1.4*10-3 1.1*10-3 1.3*10-3
From
A
T G

0.42 0.25 0.16
G C G
G
0.11 0.25 0.5
C
0.14 0.25 0.2
A G A
Hidden State
Ex Ig It
0.33 0.25 0.14
T G Exon = Max( P(Ex|Ex)*Pn-1(Ex), P(Ex|Ig)*Pn-1(Ig), P(Ex|It)*Pn-1(It) ) *P(T|Ex) = 1.1*10-2 Introgenic =Max( P(Ig|Ex)*Pn-1(Ex), P(Ig|Ig)*Pn-1(Ig), P(Ig|It)*Pn-1(It) ) * P(T|Ig) = 3.5*10-3 Intron = Max( P(It|Ex)*Pn-1(Ex), P(It|Ig)*Pn-1(Ig), P(It,It)*Pn-1(It) ) * P(T|It) = 1.3*10-3
Ex 0.1 Ig 0.89 It 0.01


From
A
T G
2.4*10-3
4.3*10-4
2.9*10-4

0.42 0.25 0.16
G C G
G
0.11 0.25 0.5
C
0.14 0.25 0.2
A G A
Hidden State
Ex Ig It
0.33 0.25 0.14
T G
Ex 0.1 Ig 0.89 It 0.01 Exon = Max( P(Ex|Ex)*Pn-1(Ex), P(Ex|Ig)*Pn-1(Ig), P(Ex|It)*Pn-1(It) ) *P(T|Ex) Introgenic =Max( P(Ig|Ex)*Pn-1(Ex), P(Ig|Ig)*Pn-1(Ig), P(Ig|It)*Pn-1(It) ) * P(T|Ig) Intron = Max( P(It|Ex)*Pn-1(Ex), P(It|Ig)*Pn-1(Ig), P(It,It)*Pn-1(It) ) * P(T|It)


From
A
T G
2.4*10-3
7.2*10-4
4.3*10-4
6.1*10-5
2.9*10-4
7.8*10-5

0.42 0.25 0.16
G C G
G
0.11 0.25 0.5
C
0.14 0.25 0.2
A G A
Hidden State
Ex Ig It
0.33 0.25 0.14
T G


From
A
T G
2.4*10-3
7.2*10-4 5.5*10-5
4.3*10-4
6.1*10-5 1.8*10-5
2.9*10-4
7.8*10-5 7.2*10-5

0.42 0.25 0.16
G C G
G
0.11 0.25 0.5
C
0.14 0.25 0.2
A G A
Hidden State
Ex Ig It
0.33 0.25 0.14
T G


From
A
T G
2.4*10-3
7.2*10-4 5.5*10-5 4.3*10-6
4.3*10-4
6.1*10-5 1.8*10-5 2.2*10-6
2.9*10-4
7.8*10-5 7.2*10-5 2.9*10-5

0.42 0.25 0.16
G C G
G
0.11 0.25 0.5
C
0.14 0.25 0.2
A G A
Hidden State
Ex Ig It
0.33 0.25 0.14
T G


From
A
T G
2.4*10-3
7.2*10-4 5.5*10-5 4.3*10-6 7.2*10-7 9.1*10-8 1.1*10-7 8.4*-9 4.9*-9 1.4*10-9 1.1*10-10
4.3*10-4
6.1*10-5 1.8*10-5 2.2*10-6 2.8*10-7 3.5*10-8 9.1*10-9 2.7*10-9 4.1*10-10 1.2*10-10 3.6*10-11
2.9*10-4
7.8*10-5 7.2*10-5 2.9*10-5 4.6*10-6 1.8*10-6 2.0*10-7 8.2*10-8 9.2*10-9 1.2*10-9 4.7*10-10

0.42 0.25 0.16
G C G
G
0.11 0.25 0.5
C
0.14 0.25 0.2
A G A
Hidden State
Ex Ig It
0.33 0.25 0.14
T G
Hidden Markov Models

How to Train an HMM
The forward-backward algorithm
Ugly probability theory math:
Starts with an initial guess of parameters Refines parameters by attempting to reduce the errors it provokes with fitted to the data.
Normalized probability of the Forward probability of arriving at the state given the observable cross multiplied by the backward probability of generating that observable given the parameter.
The Algorithms
Genetic Algorithms
Individuals are series of bits which represent candidate solutions
Functions Structures Images Code
Based on Darwin evolution

individuals mate, mutate, and are selected based on a Fitness Function
Genetic Algorithms
Encoding Rules
Gray bit encoding
Bit distance proportional to value distance
Selection Rules
Digital / Analog Threshold Linear Amplification Vs Weighted Amplification
Mating Rules
Mutation parameters Recombination parameters
Genetic Algorithms
When are they useful?
Movements in sequence space are funnel shaped with fitness function
Systems where evolution actually applies!
Examples
Medicinal chemistry Protein folding Amino acid substitutions Membrane trafficking modeling Ecological simulations Linear Programming Traveling salesman
The Algorithms
Neural Networks
1943 McCulloch and Pitts Model of how Neurons process information
Field immediately splits
Studying brains
Neurology
Studying artificial intelligence

Neural Networks
Neural Networks: A Neuron, Node, or Unit

Wa,c
W0,c
(Bias)
Activation (W)- W0,c Function a z (Bias)
Wc,n
Output
Wb,c
Neural Networks: Activation Functions Sigmoid Function Threshold Function

(logistic function) +1 out +1 out
In Zero point set by bias
In
Threshold Functions can make Logic Gates with Neurons!

Wa,c = 1
Logical And
B
W0,c = 1.5
(Bias)
(W)- W0,c
a z
(Bias)
Output
1 0
1 1 0
0 0 0
Wb,c = 1
If ( (w) Wo,c > 0 ) Then FIRE Else
Dont
And Gate: Trace

Off
Wa,c = 1
W0,c = 1.5
(Bias)
-1.5
-1.5 < 0
Off
Off
Wb,c = 1
And Gate: Trace

On
Wa,c = 1
W0,c = 1.5
(Bias)
-0.5
-0.5 < 0
Off
Off
Wb,c = 1
And Gate: Trace

Off
Wa,c = 1
W0,c = 1.5
(Bias)
-0.5
-0.5 < 0
Off
On
Wb,c = 1
And Gate: Trace

On
Wa,c = 1
W0,c = 1.5
(Bias)
0.5
0.5 > 0
On
On
Wb,c = 1

Wa,c = 1
W0,c = 0.5
Logical Or B
(Bias)
(W)- W0,c
a z
(Bias)
U 1 A 0
1 1 1
0 1 0
Wb,c = 1
Dont
Or Gate: Trace
Off
Wa,c = 1
W0,c = 0.5
(Bias)
-0.5
-0.5 < 0
Off
Off
Wb,c = 1
Or Gate: Trace
On
Wa,c = 1
W0,c = 0.5
(Bias)
0.5
0.5 > 0
On
Off
Wb,c = 1
Or Gate: Trace
Off
Wa,c = 1
W0,c = 0.5
(Bias)
0.5
0.5 > 0
On
On
Wb,c = 1
Or Gate: Trace
On
Wa,c = 1
W0,c = 0.5
(Bias)
1.5
1.5 > 0
On
On
Wb,c = 1

Wa,c = -1
Logical Not
W0,c = -0.5
(Bias)
(W)- W0,c
a z
(Bias)
1 ! 0
0 1
Dont
Not Gate: Trace

Off
Wa,c = -1
W0,c = -0.5
(Bias)
-0.5
0.5 > 0
On
0 (-0.5) = 0.5
Not Gate: Trace

On
Wa,c = -1
W0,c = -0.5
(Bias)
-0.5
-0.5 < 0
Off
-1 (-0.5) = -0.5
Feed-Forward Vs. Recurrent Networks

Feed-Forward
No Cyclic connections Function of its current inputs No internal state other then weights of connections
Out of time
Recurrent
Cyclic connections Dynamic behavior
Stable Oscillatory Chaotic
Response depends on current state

In time
Short term memory!
Feed-Forward Networks
Knowledge is represented by weight on edges
Modeless!
Learning consists of adjusting weights Customary Arrangements

One Boolean output for each value Arranged in Layers
Layer 1 = inputs Layer 2 to (n-1) = Hidden Layer N = outputs
Perceptron 2 layer Feed-Forward network
Layers
Input
Output
Hidden layer
Perceptron Learning
Gradient Decent used to reduce error
Essentially:
New Weight = Old Weight + adjustment Adjustment = X error X input X d(activation function)
= Learning Rate
Hidden Network Learning

Back-Propagation
Essentially:
Start with Gradient Decent from output Assign blame to inputting neurons proportional to their weights Adjust weights at previous level using Gradient decent based on blame
They dont get it either: Issues that arent well understood

(Learning Rate) Depth of network (number of layers) Size of hidden layers
Overfitting Cross-validation
Minimum connectivity
Optimal Brain Damage Algorithm
No extractable model!
How Are Neural Nets Different From My Brain?

1. Neural nets are feed forward
Brains can be recurrent with feedback loops
2. Neural nets do not distinguish between + or connections

In brains excitatory and inhibitory neurons have different properties
Inhibitory neurons short-distance
Frasers Rules
3. Neural nets exist Out of time

Our brains clearly do exist in time We have very little idea how our brains are learning
4. Neural nets learn VERY differently

In theory one can, of course, implement biologically realistic neural networks, but this is a mammoth task. All kinds of details have to be gotten right, or you end up with a network that completely decays to unconnectedness, or one that ramps up its connections until it basically has a seizure.
Frontiers in AI
Applications of current algorithms New algorithms for determining parameters from training data
Backward-Forward Backpropagation
Better classification of the mysteries of neural networks Pathology modeling in neural networks Evolutionary modeling

Artificial Intelligence and Learning Algorithms: Presented by Brian M. Frezza 12/1/05

Uploaded by

Copyright:

Available Formats

Artificial Intelligence and Learning Algorithms: Presented by Brian M. Frezza 12/1/05

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Artificial Intelligence and Learning Algorithms: Presented by Brian M. Frezza 12/1/05

Uploaded by

Copyright:

Available Formats

Artificial Intelligence and Learning Algorithms

Presented By Brian M. Frezza 12/1/05

Artificial Neural Networks Vs Neuron Biology

Whats a Learning Algorithm?

Training Data Test Data

Network inference Classification inference

Analogies To Chemical Biology

Fundamental Theories of Intelligence

Subcellular location through fluorescence

Bayesian Networks Hidden Markov Models Genetic Algorithms Neural Networks

Bayesian Networks: Basics

Predicts as a weighted average

Bayesian Network Example

Bayesian Network: Trace

Prediction: Will their next kid be a Redhead?

= P(red|Ha)*P(Ha) + P(red|Hb)*P(Hb) + P(red|Hc)*P(Hc) = (1)*(1/3) + (1/2)*(1/3) + (0)(1/3) =(1/2)

Prediction: Will their next kid be a Redhead?

= P(red|Ha)*P(Ha) + P(red|Hb)*P(Hb) + P(red|Hc)*P(Hc) = (1)*(1/2) + (1/2)*(1/2) + (0)(1/3) =(3/4)

Bayesian Network: Trace

Prediction: Will their next kid be a Redhead?

= P(red|Ha)*P(Ha) + P(red|Hb)*P(Hb) + P(red|Hc)*P(Hc) = (1)*(3/4) + (1/2)*(1/4) + (0)(1/3) =(7/8)

Bayesian Network: Trace

Prediction: Will their next kid be a Redhead?

= P(red|Ha)*P(Ha) + P(red|Hb)*P(Hb) + P(red|Hc)*P(Hc) = (1)*(7/8) + (1/2)*(1/8) + (0)(1/3) =(15/16)

Bayesian Networks Notes

Programmer needs to form hypothesis beforehand.

Bayesian Networks Hidden Markov Models Genetic Algorithms Neural Networks

Hidden Markov Models(HMM)

Hidden Markov Models: Take a Step Back

1st order Markov Model Setup

For each time step

1st order Markov Model Trace

1st order Markov Model Trace

1st order Markov Model Trace

1st order Markov Model Trace

Im going to stop here. Markov Chain:

What else can Markov do?

Continuous Markov Models

Hidden Markov Models

Hidden Markov Models (HMMs)

Secondary structure prediction for protein

Hidden Markov Models: Example

Hidden Markov Models: Smaller Example Exon/Intron Mapping

Hidden Markov Models: Smaller Example Exon/Intron Mapping

Ex Ig It 0.7 0.1 0.2 0.49 0.5 0.01 0.18 0.02 0.8

Ex 0.33 0.42 0.11 0.14 Ig 0.25 0.25 0.25 0.25 It

Hidden Markov Model

Viterbi Algorithm: Trace

Example Sequence: ATAATGGCGAGTG

Observable State Probabilities Observable A T

0.33 0.25 0.14

Viterbi Algorithm: Trace

Example Sequence: ATAATGGCGAGTG

Observable State Probabilities Observable A T

0.33 0.25 0.14

Viterbi Algorithm: Trace

Example Sequence: ATAATGGCGAGTG

Observable State Probabilities Observable A T

0.33 0.25 0.14

= P(red|Ha)P(Ha) + P(red|Hb)P(Hb) + P(red|Hc)P(Hc) = (1)(1/3) + (1/2)*(1/3) + (0)(1/3) =(1/2)

= P(red|Ha)P(Ha) + P(red|Hb)P(Hb) + P(red|Hc)P(Hc) = (1)(1/2) + (1/2)*(1/2) + (0)(1/3) =(3/4)

= P(red|Ha)P(Ha) + P(red|Hb)P(Hb) + P(red|Hc)P(Hc) = (1)(3/4) + (1/2)*(1/4) + (0)(1/3) =(7/8)

= P(red|Ha)P(Ha) + P(red|Hb)P(Hb) + P(red|Hc)P(Hc) = (1)(7/8) + (1/2)*(1/8) + (0)(1/3) =(15/16)