Spike Triggered Covariance in Strongly Correlated Gaussian Stimuli
Figure 1
Spike triggered covariance analysis of simulated spike trains of a model with a single feature orthogonal to the coherent mode.
(A) Inputs are plotted when projected on the eigenvectors corresponding to the largest eigenvalues of (in units of pixel illumination). Marginal distributions are plotted for each dimension. Inputs that elicited a spike are shown in red, and those that did not in blue. By construction, the change in variance is larger along the first dimension. (B) The empirical eigenvalue distribution of
(black) compared to the null distribution (blue). No eigenvalues of
are found to be significant (shaded area indicates
% confidence intervals for the support of the null distribution) (C) Rank ordered eigenvalues (black) plotted with the null distribution (blue). (D) Nested rank-wise significance testing. The highest ranked eigenvalues of
are within the
% confidence intervals derived from the null distribution constructed for each rank separately (see Materials and Methods for details). (E) For each random spike train we computed
, the variance of the projection of the spike-triggered stimulus on the relevant feature
(distribution shown in yellow). The purple line indicates
for the real spike train, suggesting the spike train contains enough signal to determine the relevant feature as significant. Inset shows the relevant feature, a
image patch (
). Simulation details:
,
,
repetitions to find
.