Figure 8.

Scatterplot showing the emergence of polychronous spatio-temporal structure in neuronal spike times after training and through successive network layers. Each marker corresponds to an individual neuron in either of the higher layers 1 or 2 that was activated by the stimulus across all test presentations. The following four sets of simulation results are presented: pretraining layer 1 neurons (blue dots), pretraining layer 2 neurons (orange crosses), post-training layer 1 neurons (green dots) and post-training layer 2 neurons (red crosses). For each neuron, the mean time of its first spike across all 10 simulations in which a stimulus is presented (abscissa) is plotted against the standard deviation in these first spike times (ordinate). In these simulations, the axonal transmission delays between the input layer and layer 1 are uniformly distributed between 1 and 10 ms, while axonal delays between layer 1 and layer 2 are uniformly distributed between 1 and 30 ms. It is evident that training the network leads to a significant reduction in the standard deviations of first spike times in layers 1 and 2. Thus, training the network using STDP reduces the degree of temporal variation in the first spike times. Moreover, layer 2 neurons have reduced standard deviations in their first spike times compared with layer 1 both before and after training. So successive layers of processing also reduce the degree of temporal variation in the first spike times as hypothesized. Copyright © 2018 American Psychological Association. Reproduced [or Adapted] with permission. The official citation that should be used in referencing this material is [9]. No further reproduction or distribution is permitted without written permission from the American Psychological Association.