Business Systems and Knowledge Modelling Lab

Media Release

Uni team probes alarm patterns

Just minutes after a power line goes down, a board in the SECV's Thomastown terminal lights up like a Christmas tree much to the dismay of systems operators on the job.

This single event can trigger thousands of alarms - bamboozling staff who must try to establish the origin of the problem.

Technologists at La Trobe University have produced a prototype now being used by the SECV to interpret these conplex alarm patterns and give operators an intelligent assessment of the situation.

La Trobe is leading the international circuit with project work that emulates the human ability to deal with complex problems, especially those which need to be responded to quickly.

Project leader Professor Tharam Dillon has been involved in numerous international conferences - as a speaker and organiser - on the subject of neural and symbolic systems. He believes his team is one of only threee in the world working to combine the two disciplines for greater effect in intelligent problem solving.

The first system, called BRAINNE (Building Representations for Artificial Intelligence Using Neural Networks), uses neural networks as the basis for learning symbolic artificial representation and has been applied to databases as diverse as mushroom and soya bean recognition, sonar data and United States congressional voting patterns.

The Royal guide Dogs Association is working with La Trobe University to develop a sonar pathfinder for people with impaired sight, using BRAINNE.

The second system is GENUES (Generic Neural Expert System) which combines symbolic artificial intelligence structures with neural networks in reasoning and problem solving.

This is the system which has been applied to the SECV's alarm dilema.

Professor Dillon explains the concept of the integrated neural network with symbolic artificial intelligence using the analogy of a five-year-old child learning how to hit a cricket ball. A child is unlikely to be able to explain in natural languge how he is capable of hitting the ball when it is pitched short, but he is obviously able to recognise patterns of the balls that are thrown to him and develop patterns of response accordingly. So there is a degree of neural processing going on but it is sub-symbolic because he cannot communicate it in natural languge.

An older child may learn from his coach that when a series of if-then rules are applied, he may improve his performance in hitting the ball. This knowledge is symbolic but not formal. Professor Dillon said it was these qualities that researchers at La Trobe were trying to emulate in their projects.

- The Age, 1995