Figure 4.

Triplets are better than pairs when the neuron adapts. (A) Distributions of synaptic efficacies obtained after learning. The weights were all initialized at 1 mV before learning (black arrow). When SFA is switched off, the very same bimodal distributions emerge (not shown). (B) Evolution of the MI along learning time. Learning time is arbitrarily indexed from 0 < α < 1. The dashed curves represent the MI when the weights taken from the momentary distribution at time α are shuffled. Each point is obtained from averaging the MI over 10 different shuffled versions of the synaptic weights. Error bars denote standard error of the mean (SEM) over 10 independent learning episodes with different input spike trains. (C) Same as in (B), but SFA is switched off. The y-scale is the same as in (B). Parameters for those simulations were λ = 0, γ = 0 with SFA, and γ = 1 without SFA. Other parameters took the values given in Table 1.