Fig 2.

(A) Two ensembles of spike patterns are generated by distinct, N-dimensional inputs I^A and I^B repeatedly presented to the network on multiple trials, where initial states are chosen at random. This results in two network-wide response ensembles Θ_I^A and Θ_I^B containing spike patterns across neurons, time and trials. (B) Illustration of the two procedures used to control input similarity. Left: ρ_same controls the number of neurons N_same = ρ_same N that receive identical inputs I_i(t) under both stimuli I^A(t) and I^B(t). Right: correlation coefficient ρ_corr controls the correlation of all pairs of neural inputs $I_i^A\left(t\right)$ and $I_i^B\left(t\right)$ (note that I_i(t) and I_j(t) remain uncorrelated if i ≠ j).