Figure 3.

This shows the development of the SKIM method for a spatio-temporal spike pattern recognition system. The structure of one dendritic branch and the soma is shown at the top. The signals are, from top: the input pattern on five channels and the target output; a test sequence with added noise spikes, and target output; the summed spike output from one dendritic branch; the resulting non-linearly transformed alpha function output from that branch; the soma potential for the output neuron; and the resulting output spike train. It can be seen that the spike pattern is successfully recognized in the presence of some spike noise. The target output sequence has spikes which have been widened to 10 timesteps in order to increase the non-zero target energy, as described in the text. Note that where input patterns overlap closely, the output spikes may merge together and appear as one event, as seen at 630 and 850 ms.