Artificial Neural Networks ECE.09.454/ECE.09.560 Fall 2006

1
Artificial Neural Networks ECE.09.454/ECE.09.560
Fall 2006
Lecture 9November 13, 2006

• Shreekanth Mandayam
• ECE Department
• Rowan University
• http//engineering.rowan.edu/shreek/fall06/ann/

2
Plan

• Unsupervised Learning Other Neural Net
  Architectures
• Self Organizing Maps (SOMs)
• Kohonen Network
• Recurrent Networks
• Hopfield Network
• Final Project Discussion

3
Classical ANN Paradigm
Stage 1 Network Training
Feedforward Artificial Neural Net
Present Examples
Stage 2 Network Testing
Feedforward Artificial Neural Net
New Data
4
What if?

• Desired outputs are unknown
• Input data is partially complete
• Neural net is not just feedforward

Unsupervised Learning
Self-organizing Maps Recurrent
Networks
5
Self Organizing Maps
wi
xi i(xi)

• Location of the winning neuron is based upon the
  class of the input signal
• Similar input signals map on to winning neurons
  that are located close to each other
• The location and synaptic weights are determined
  using neuron
• Competition
• Cooperation
• Adaptation

Matlab Demos Competitive learning 2-D Self
organizing map
6
Recurrent NetworksWhat if?
Content Addressable Memory (CAM)
7
Content Addressable Memory
The input x is stored in the equilibriumneuron
states
x
The network falls into the appropriate equilibr
ium state
Perturbed/partial x
Matlab Demo Hopfield 2-Neuron
8
Final Project Discussion

• Topics and Dates

9
Lab 3 RBF Neural Nets

• http//engineering.rowan.edu/shreek/fall06/ann/la
  b3.html

10
Summary