Figure 1.

Three step process describing a reservoir computing model extended by having the recurrent connections adapted with unsupervised plasticity in a pre-training phase. Firstly, input samples I are presented in random order while the resulting neural activity drives synaptic adaptation under plasticity. Secondly, each input sample is presented in sequence with the resulting neural activity decoded into a series of state vectors S. Finally, the state vectors are used as the input to train a set of perceptron readouts, one to recognize each class of sample, Cx.