Reservoir Computing for Sensory Prediction and Classification in Adaptive Agents

Artificial neural networks are an in silico laboratory for studying the dynamics of the brain. In recurrent networks, the units’ activations are recurrently fed back into the network. Thereby complex network dynamics emerge that extend over longer time scales than the individual units’ activation time constants. The recurrent echo-state networks with their fixed connection weights acquire an internal representation that uniquely depends on the input history, but not on the initial state of the network. We present echo-state networks as models of sensory systems and sketch two examples of their usage in learning agents. The first example is gesture classification from moving camera images, and the second is a conceptual account of timing. Furthermore, we review a recent idea of self-prediction augmenting an echo-state network. The weights self-predicting the internal state filter out external noise, and improve the network performance significantly. Together, this chapter presents exciting new developments in the field of reservoir computing.