Dependence of selectivity and sparseness on input.

<p>A) and B) Selectivity and sparseness as a function of excitatory network firing rate. Mean input rate is modulated keeping its sparseness fixed. C) Selectivity and sparseness as a function of input sparseness. Input mean firing rate adjusted so that excitatory network fires at 10 Hz at default input SPI = 0.75. To achieve less sparse input population rate distributions than the minimum sparseness allowed by our default stimulus ensemble, we set , where is the mean of , and therefore . SLI and SPI are now given by . Note that numerically calculating 0 for SLI is impossible due to finite network simulation time. Trial-to-trial variability of Poisson-like firing neurons will lead to a residual selectivity even if the input is unselective, which will vanish only in the limit of infinite simulation time. In our simulations, we increased simulation time as needed to minimize this effect, especially for low input selectivity. D) Same as in C) but synaptic weight is changed from its default value to .