Comparisons of separability and variance of evidence distributions when noise and variability are manipulated independently for the 4-layer CNN across the three experiments.

For contrast manipulations, the stimulus contrast was increased while stimulus variability was held constant. For variability manipulations, the stimulus noise or variability was increased while stimulus contrast was held constant. Comparisons using two-sided, pairwise t-tests showed that separability of evidence distributions increased significantly from the low- to high-contrast condition whereas the variance of evidence distribution decreased from the low- to high-contrast conditions for all experiments except Experiment 3. For Experiment 3, increasing contrast increased the variance of evidence distributions. On the other hand, separability of evidence distributions decreased significantly from the low to high variability condition and the variance of evidence distribution increased from the low to high variability conditions for all experiments.
(XLSX)