Fig 7.

(A) Top: Trained Tempotron weights w_i. Bottom: example cross-trial spiking output for neurons associated with large and small weights w_i. Asterisk shows the time at which the maximal value of the Tempotron variable V(t) is obtained (see also Fig 3 (B)). (B) Spike-time reliability for the closest spike to the maximum (asterisk), plotted for each cell against the readout weight w_i. Solid blue markers indicate responses to I^A(t) and hollow red markers to I^B(t). (C) Example raster plot for a single neuron across 100 trials, with and without spike-time jittering. Two jitter strengths shown: ±25 and ±100 ms. (D) The Tempotron’s performance P when all neurons are jittered, plotted against jitter strength, for three values of ρ_same. The classifier was trained and tested on jittered spike time data. (E) Tempotron’s performance P against the cumulative inclusion of neurons where the spike times were strongly jittered (±50 milliseconds). Different lines indicate three orderings of neurons the neurons that were jittered, based on their trained readout weight w_i: (i) decreasing, (ii) increasing and (iii) random. The classifier was trained on original data and tested on jittered data from a network with N = 500, K = 20. For all panels except (D), ρ_corr = ρ_same = 0.