Fig. 3: Backward masking.

A Linear decoder results of the preferred class for different delays, regardless of stimulus perception. Time on x-axis represents middle of 100 ms bins used for decoding. Lines and shading represent the mean ± standard error classification AUC from 10 decoding repetitions. Horizontal bars represent significant decoding performance compared to random label shuffling (p < 0.001, permutation test with at least five consecutive significant time points). The bars below the plots show the number of trials per condition. B Average z-normalized spike rate from all channels selective for the preferred class (preferred class > all other classes; p < 0.01, one-sided permutation test). Shading signifies standard error. Asterisks indicate significance between perceived and nonperceived target stimuli within the response window (A2 [130–280], A4 [200–350]) (*p < 0.05, **p < 0.01, one-sided independent samples t-test). C Decoding results for preferred class for perceived and nonperceived trials. Lines and shading represent the mean ± standard error classification AUC from 10 decoding repetitions. Time on x-axis represents middle of 100 ms bins used for decoding. Horizontal bars represent significantly better decoding performance for perceived stimuli (p < 0.01, one-sided independent samples t-test). The same delays are grouped as in (B). The bars in the left upper corner of each plot in (B and C) show the number of perceived (green) and nonperceived (red) trials for the preferred class (left) and all other classes (right). For array 2, the other class is scrambled images (blue).