Pattern completion in the context of moving bars across a visual field.

<p>A 400 neuron network arranged in a 20×20 sheet with topographic input from the visual field is trained on bars moving left, right, up and down (See <a href="http://www.plosone.org/article/info:doi/10.1371/journal.pone.0072865#pone.0072865.s004" target="_blank">Figure S4</a> for details). The network carries out pattern completion in response to bars moving right (top) and left (bottom). The clean input is presented with equal power random noise during the open state. The network performs perfect pattern completion during the second closed state (dashed boxes) following the presentation of a long-enough sequence of moving bars during the preceding open state. Open/closed durations are sampled from a Gaussian with a mean of 7 and sd of 2 time steps. For the sake of space, bars moving up and down are omitted and only every other time-point is shown.