CSE5301 Neuro-Fuzzy Computing


A/Prof. Andrew P. Paplinski

Prerequisite knowledge

Basic knowledge of vectors and matrices is assumed. Specialised mathematical concepts will be introduced.

Syllabus

This units examines mathematical and computational fundamentals of artificial neural networks and Fuzzy systems, and their applications in signal and image processing, pattern recognition and modelling. The syllabus includes:
Basic concepts of neurocomputing:
Artificial Neural Networks (ANN) and their biological roots and motivations. ANNs as numerical data/signal/image processing devices. Encoding (training phase) and decoding (active phase). Taxonomy of neural networks: feedforward and recurrent networks with supervised and unsupervised learning laws. Static and dynamic processing systems. Basic data structures: mapping of vector spaces, clusters, principal components.


Basic terminology related to an artificial neuron:
a summing dendrite, synapses and their weights, pre- and post-synaptic signals, activation potential and activation function. Excitatory and inhibitory synapses. The biasing input. Types of activating functions.


The Perceptron
The Perceptron and its learning law. Classification of linearly separable patterns.


Linear Networks.
Adaline --- the adaptive linear element. Linear regression. The Wiener-Hopf equation. The Least-Mean-Square (Widrow-Hoff) learning algorithm. Method of steepest descent. Adaline as a linear adaptive filter. A sequential regression algorithm.


Multi-Layer Feedforward Neural Networks:
aka Multi-Layer Perceptrons. Supervised Learning. Approximation and interpolation of functions. Radial-Basis functions. Back-Propagation Learning law. Fast training algorithms. Applications of multilayer perceptrons: Image coding, Paint-quality inspection, Nettalk.


Self-Organising systems.
Unsupervised Learning. Local learning laws. Generalised Hebbian Algorithm. The Oja's and Sanger's rules. Principal component analysis --- Karhunen-Loeve transform.


Competitive Learning:
MinNet and MaxNet networks. Clustering. Learning Vector Quantisation. Codebooks. Application in data compression.


Self-Organising Feature Maps:
Kohonen networks.


Recurrent networks
Hopfield networks.


Fuzzy logic Systems
Basic definitions and operations.
Fuzzy relations
Fuzzy rules
Fuzzy inference
Fuzzification and de-fuzzification
Adaptive Neuro-Fuzzy Inference Systems

Recommended references:


Subject structure and organisation

The subject format is based on two hours a week of lectures and two hours a week of practical/tutorial work. It is expected that in addition a student spends approximately 6 hours a week on theoretical and practical aspects related to the unit.

Practical work

Practical work related to the unit is based on the MATLAB package. MATLAB is available on many (but not all) Unix/Linux/Windows platforms around the campus. You can also purchase a MATLAB Student Version, in particular, from http://www.mathworks.com/academia/student_version/

Introduction to MATLAB is given in Practical 1.

Make sure that you have your computer account active asap.

Assessment

is based on assignments and practical works (50%) and two-hour exam (50%).

Plagiarism

Students should consult University materials on cheating, in particular:

  1. Student Resource Guide - section on Student Rights and Responsibilities at http://www.monash.edu.au/pubs/handbooks/srg/srg0059.htm
  2. Student Resource Guide at http://www.monash.edu.au/pubs/handbooks/srg/, particularly the section on Cheating at http://www.monash.edu.au/pubs/handbooks/srg/srg0071.htm

It is the student's responsibility to make themselves familiar with the contents of these documents.


Andrew P. Paplinski
25 February 2005