ALAR: Artificial Life and Adaptive Robotic Lab

Computer Science Faculty, UNSW@ADFA

Canberra, Australia

Evolutionary Artificial Neural Networks


Anti Correlation Learning

Participants: Hussein Abbass and Bob McKay

Anti-correlation has been used in training artificial neural network ensembles. Negative correlation learning (NCL) is the state of the art anti-correlation measure. Recently, Mckay and Abbass (2001a) proposed an alternative anti-correlation measure, RTQRT-NCL, which shows significant improvements on the test examples, particularly with larger ensembles. The same measure was also successfully used with genetic programming (Mckay and Abbass, 2001b).

The aim of this project is to further analyse RTQRT-NCL and improve its performance. The outcome of this research will be in terms of research publications in conferences and journals.


Evolutionary Artificial Neural Networks

Participants: Hussein Abbass

The aim of this project is to develop free-architecture evolutionary neural networks models. We use techniques from multiobjective and memetics to evolve artificial neural networks. We are also interested in evolving neural networks controllers.