Figure 10.

Learning of more complex feature detectors, showing the output of the system when presented with stimuli composed of digits from 1 to 9. Each snapshot on the top left shows the analog input to the ESN, obtained by filtering the DVS events. The top right shows the prediction of the best ESN, as selected by the WTA circuit. Below, the current predictions of all nine ESNs used in the experiment are shown. A white square around the prediction indicates the ESN that was selected by the WTA. The snapshots show the progression of learning, starting with an untrained network in (A), which only produces random predictions. (B–D) show the output of the same networks after the presentation of patterns “1,” “2,” and “3,” (respectively). A white mark underneath an ESN prediction indicates that this ESN has learned a feature. Finally (E) shows its output after the end of the learning process where all networks have learnt an input stimulus.