Peter Ford Dominey
In reservoir computing, inputs are presented to a network of neurons with fixed recurrent connections (the reservoir). The recurrent connections yeild a dynamic system that generates a high dimensional mix in space and time of the input sequence. Modifiable connections to readout neurons are trained to generate the desired function of the input. First developed to describe prefrontal cortical neural activity during behavioral sequence learning, reservoirs have been used for advanced signal processing and to simulate neural activity in cortex. Recently, accumulating data indicate that the primate cortex behaves as a reservoir in its information processing capabilities. This symposium will address (a) recent findings that link reservoir activity to cortical activity in the behaving primate, and (b) how reservoirs can be used in cognitive architectures for higher cognitive functions like language processing.