COMP9444/9844 Neural Networks

Feedback Quiz on Self-organising Systems - Solutions

It is best not read the answers until you've tried to answer the questions yourself.

1. What is the difference between supervised learning and unsupervised learning?

   Answer: In supervised learning, training patterns giving inputs and the corresponding correct outputs are available, while in unsupervised learning, the system must find interesting and/or significant patterns in the data without any feedback as to what is "right".

2. In a self-organising map, how are the output neurons organised?

   Answer: into an array, whether 1-D, or 2-D, or 3-D or beyond. Being near to something in the sense of the arrays natural ordering is important in the SOM.

3. What is the name of the process in Kohonen's SOM algoritm in which the winning neuron is chosen?

   Answer: The competitive process.

4. If a neuron is next to the winning neuron in a cycle of the SOM learning algorithm, in the middle of the self-organising or ordering phase, will its weights change?

   Answer: Yes - the closer you are to the winning neuron, the more your weights change.

5. In the neighbourhood width function σ(n) described in lectures, does the width increase or decrease as time goes on, and is the change linear, geometrical or exponential?

   Answer: It decreases geometrically.

6. What is the "diameter" of a m×n 2-D SOM?

   Answer: max(m, n)

7. What is a quantizer?

   Answer: A quantizer is an algorithm that finds, for each input vector, a nearby vector to represent the input vector. The quantizer then outputs a compact label that corresponds the representation vector (which is also known as a reconstruction vector).

Bill Wilson's contact info

UNSW's CRICOS Provider No. is 00098G