Robustness of iSTDP learning rule.
The robustness of the iSTDP learning rule for stochastic spiking inputs was checked by comparing the learning rates of spiking simulations with the theoretical predictions from equation 7. Effective learning rate is plotted as a function of frequency for the iSTDP profile constrained by the experimental data shown in Fig. 3H. Learning rate was calculated for sinusoidally modulated Poisson spike trains using the procedure described in Fig. 8 for in-phase sinusoidal modulation frequencies in the range 0.05 to 50 Hz. The blue curve is for spike trains with tonic rate 60 spikes/sec and a modulation depth of 40 spikes (summarised as 60±40). The theoretical (dashed) learning rate curve (also shown in Fig. 2B) is overlaid for comparison. The green curve (60±20) and red (30±20) learning rate curves are calculated using different values for tonic and for both tonic and peak modulation firing rates respectively. The results illustrate the theoretical predictions that (i) peak learning rate is proportional to the product of modulation rates (in this case both peak modulations are halved, reducing peak learning rate to 25% of its value) (ii) the overall shape (frequency dependence) of the learning curve is unaffected by modulation level, and (iii) the tonic rate has no effect on learning.