Figure 13.

State-dependent computing in the reservoir layer is affected by new training data. (A) The structural network connected to readout units #4 and #8 after 1 h of training. These readout units that are maximally active during the presentation of the new pattern #26 for d = 1.4 s. (B) The structural network for the combined training data of old and new input patterns after 2 h of training is shown here. Synaptic plasticity alters the structural network as expected. (C) The transitions between E → E neurons in the reservoir after 1 h of training This functional network shows firing activity at many more neurons and this is manifested by the novelty of the new pattern that the network has never processed before. (D) This functional network after 2 h of training is sparser compared to the one shown after 1 h since the network is able to learn the new pattern by adapting its receptive fields and generating new readout codes that are more discriminatory.