CSC437 Neural network fundamentals

Course outline:
This units examines formal fundamentals of neural networks and their basic applications, typically related to image processing. The syllabus includes: Basic concepts of neurocomputing. Single-layer feedforward neural networks. Encoding and decoding -- the relaxation phase and the training phase. The Perceptron and its learning law. Pattern classification. Linear Networks -- Adaline. The Least-Mean-Square algorithm. Method of steepest descent. Multi-Layer Feedforward Networks (Peceptrons) with Supervised Learning. Approximation of functions. Back-Propagation Learning law. Fast training algorithms. Radial-Basis function networks and their application in function interpolation and approximation. Unsupervised Hebbian Learning. Principal component analysis. Self-Organizing systems. Competitive Learning. Learning Vector Quantization. Self-Organizing Feature Maps. Recurrent neural networks. Hopfield networks. Associative memories. The Boltzman machine. Recurrent backpropagation Neural networks. Adaptive Resonance (ART) memory.

Lectures:
Wednesday 2.00-4.00pm (S15)

Assessment: Assessment will be based on reports/assignments submitted as a result of practical work.

Practical work and related assignments
PracAssignment 1 (10%) due Wednesday 15 March (week 3) Changed to Wednesday 22 March (week 4)
PracAssignment 2 (20%) due Wednesday 29 March (week 5)
PracAssignment 3 (20%) due Wednesday 12 April (week 7)
PracAssignment 4 (20%) due Wednesday 3 May (week 9)
PracAssignment 5 (15%) due Wednesday 17 May Changed to Thursday 18 May (week 11)
PracAssignment 6 (20%) due Wednesday 31 May (week 13)

All assignments are due at 12 noon to the Enquiries Office.

Marking policy:
A student's work graded from "unsatisfactory" to "high distinction" attracts typically up to

85% of the total mark, that is, 13/15 or 17/20.

Higher marks can only be obtain for an extraordinary achievments.

Courseware

Back to 2000 Honours subjects