The work within the Neural Systems Engineering theme is directed at the efficient exploitation of Artificial Neural Network technology in a variety of different practical problem domains including character recognition and automated robot control. Presently, there are three major strands to the work.
The group has worked extensively to exploit the advantages offered by exploring the areas of Artificial Neural Systems and Formal Mathematical Logic in order to produce a system which is capable of generating generalised solutions to given problem domains. The idea relies on enhancing the current state of neural network technology by generalising the concept of weight and activation function employed within the technology. The system addresses the very serious problems encountered by Software Engineers in the production of formally correct and reliable software systems and represents a novel exploitation of two emerging technologies.
Major work is currently underway in developing neural systems based on direct simulation of biological neural structures. Specifically, a computational paradigm is being developed based on principles found within the operation of the mammalian visual cortex. The study has the objectives of the practical development of a novel computational paradigm supporting the artificial processing of visual scenes. These may be exploited in numerous pattern recognition and visual processing domains. The development of improved learning algorithms for artificial neural systems which will provide novel theoretical models of their operation.
The third major strand is that of Weightless RAM-based neural networks. Such networks offer advantages over conventional neural systems including efficient learning algorithms (often one-shot learning), arbitrary mappings from inputs to outputs and easier direct hardware implementation. Many network architectures based on this paradigm have been developed by the group and work is continuing both on improving these network architectures and on the design of further novel architectures. Current work is centred on creating a multi-expert robot guidance system exploiting Genetic Algorithm (GA) technology to increase the efficiency of the networks. This represents a merging of many research interests within the group.