Professional Documents
Culture Documents
Introduction
My whereabouts
• Dr. M. Castellani
• Centria – Centre for Artificial Intelligence
• Room 2.50
• Ext. 10757
• Email: mcas@fct.unl.pt
• Research Interests
– Machine Learning, Approximate Reasoning
– Evolutionary techniques for generation of Fuzzy Logic Systems
– Evolutionary techniques for generation of Neural Networks Systems
– Spiking Neural Network models.
– Data Classification, Time Series Modelling and Control.
• Introduction to ANNs
– Taxonomy
– Features
– Learning
– Applications
I
• Introduction to ANNs
– Processing elements (neurons)
– Architecture
• Functional Taxonomy of ANNs
• Structural Taxonomy of ANNs
• Features
• Learning Paradigms
• Applications
40 mV membrane
rest
20
activation
0
Action potential
-20
Action potential ≈ 100mV
-40
Activation threshold ≈ 20-30mV
-60 Rest potential ≈ -65mV
Spike time ≈ 1-2ms
-80 Refractory time ≈ 10-20ms
Refractory time
-100
ms
-120
0 10 20 30 40 50 60 70 80 90 100
x1(t)
Stimulus wi1
ui t wij x j t x2(t)
j wi2
wi3
x3(t) w ij x j (t ) yi f (ui (t)) yi(t)
wi4
j
Response
x4(t) wi5
yi t f urest ui t
x5(t)
Neuron i
1.2
output Stimulus Response
yi f urest ui
1
0.8
on ui wij x j
0.6
j
0.4
0.2
input
0
-0.2 -10
“Hard” threshold
-8 -6 -4 -2 0 2 4 6 8 10
-0.4
heaviside
-0.6
z ON
-0.8
-1
off
f z
-1.2 else OFF = threshold
output
1.2
Stimulus Response
1 on
0.8
ui wij x j yi f urest ui
0.6
j
0.4
0.2
input
0
-10 -8 -6 -4 -2 0 2 4 6 8 10
-0.2
-0.4
2/(1+exp(-x))-1 “Soft” threshold
-0.6
-0.8
2
-1
off
f z 1
-1.2 1 e z
yi (t ) f urest (t t f ) 0 t , ui
t
dz
z & 0 ON
f z dt
else OFF
1.5
0.5 t
0
0 10 20 30 40 50 60 70 80 90 100
-0.5
-1
Output layer
fully connected
Hidden layers
Input layer
sparsely connected
• Feedback connections
• Dynamic memory:
y(t+1)=f(x(τ),y(τ),s(τ)) τ(t,t-1,...)
• Jordan/Elman ANNs
• Hopfield
• Adaptive Resonance Theory (ART)
• Learning
• Approximate reasoning
• Generalisation capability
• Noise filtering
• Parallel processing
• Distributed knowledge base
• Fault tolerance
• Supervised learning
– Classification
– Control
– Function approximation
– Associative memory
• Unsupervised learning
– Clustering
• Reinforcement learning
– Control
30/09/22 Artificial Neural Networks - I 19
Supervised Learning
F yi x j F yi x j wij
d
wij F wij , yi , x j d d
dt wij yi x j wij yi x j wij
dt dt
=learning coefficient
wij=connection from neuron xj to yi y1
• Sequential tasks
• Desired action may not be known
• Critic evaluation of ANN behaviour
• Weights adjusted according to critic
• May require credit assignment
• Population-based learning
– Evolutionary Algorithms
– Swarming Techniques
– Immune Networks
30/09/22 Artificial Neural Networks - I 24
ANN Summary
Feedforward Recurrent
• Classification
• Clustering
• Associative memory
• Control
• Function approximation
• Learning capability
• Statistical classifier systems
• Data driven
• Generalisation capability
• Handle and filter large input data
• Reconstruct noisy and incomplete
patterns
• Classification rules not transparent
30/09/22 Artificial Neural Networks - I 27
Applications for ANN
Classifiers
• Pattern recognition
– Industrial inspection
– Fault diagnosis
– Image recognition
– Target recognition
– Speech recognition
– Natural language processing
• Character recognition
– Handwriting recognition
– Automatic text-to-speech conversion
• Stimulus-response association
• Auto-associative memory
• Content addressable memory
• Fast parallel distributed processing
• Robust to noise and incomplete
patterns
• Limited storage capability
• Learning/adaptation capability
• Data driven
• Non-linear mapping
• Fast response
• Fault tolerance
• Generalisation capability
• Handle and filter large input data
• Reconstruct noisy and incomplete patterns
• Control rules not transparent
• Learning may be problematic
• ANN controller
• conventional controller + ANN for
unknown or non-linear dynamics
• Indirect control schemes
– ANN models direct plant dynamics
– ANN models inverse plant dynamics