Business Systems and Knowledge Modelling Lab

Media Release

Brain game: La Trobe sets out to mimic the mind

For all those comforted by the thought that "no computer can ever replicate the processes of the human brain", the day of judgement may be closer than you think.

Researchers from La Trobe University's school of computer science and computer engineering have developed a computer that mimics the way the brain selects its knowledge.

Creating machines that work in the same way as human thinking has been a difficult problem for computer boffins.

Professor Tharam Dillon, head of the La Trobe school, and his team began thinking about the way humans learn, he told a research journal.

First, there is sub-symbolic knowledge - what we know innately but are unable to articulate. A child may pick up a ball game without instruction, but not be able to tell what they do when they perform a certain skill that is part of the game.

Computers keep such information in their neural networks, but it is often difficult to extract, depending on what inputs are fed in. Second, the symbolic knowledge is represented in the brain by the rules we use to pull out and present our knowledge. In intelligence systems, the rules are known as production rules and concept hierachies.

Finally, there is the knowledge of the mathematician or logician, presented in symbols.

Professor Dillon says humans are very good at learning from examples and patterns and then extracting knowledge in symbolic form.

The La Trobe team has developed a computer that processes knowledge into its neural networks yet, importantly, can extract the knowledge in the form of rules of thumb as well as concept hierachies. In that, it seeks to mimic the human brain.

Professor Dillon and lecturer Dr Rajiv Khosla also have been able to produce a system that can reason using neural networks and symbolic knowledge together.

- The Age, 1995