T.G. Clarkson, D. Gorse and J.G. Taylor
Synopsis
A model of neural processing is described which is able to incorporate a great deal of neurophysiological detail (synaptic noise, synapse/synapse interactions, cell surface geometry, temporal features) and which is capable of hardware realisation as a 'probabilistic random access memory' (pRAM). The model can operate either in a binary mode or can integrate the effects of incoming spike trains to perform a real-to-binary mapping ('integrating pRAM' or i-pRAM). pRAMs can be trained using a variety of techniques, in particular reinforcement training, which has the advantage of being wholly realisable in hardware as well as having greater biological plausibility than supervised techniques.
Key words
Neural modelling, pRAMs, hardware implementation, noise, reinforcement training, automata.