Implementation of a General Model of Cooperation between Cortical Areas on a Parallel System Based on Transputers

The computer implementation of a neural network of the cortical type can take advantage of the new technologies of parallel processing, in particular the systems based on transputers: (1) the neural network operations are naturally parallel; (2) the topology of the parallel system can be derived of the known anatomic structure of the cortical network, which enables the creation of a large number of relations with a limited number of local connections; the processing power as well as the memory capacity required by the succesive learning trials can be increased easily. We have designed a general software for the implementation of neural networks of biological type on transputers network, and we have developed it on a 24 transputers system. It enables us to map multilayered networks, with several communications streams on each cortical area; we can therefore evaluate for each cortical function, the relative importance of the different types of interaction between the main streams of information. We have used this system to simulate cooperation between cortical areas implied in visual recognition and command of arm movements toward visual targets.