Model MEG RSA latency.

The mean latency and 95% one-sided confidence intervals (milleseconds) for the source-estimated MEG-model RDM correlation time series for AlexNet Conv1, AlexNet FC7, and memorability magnitude model RDMs were computed using a jackknife-based relative baseline criteria. For each correlation time series, the standard deviation of the 15 subject grand-averaged baseline values was first computed (from −200–0 ms before stimulus onset). Then, for each of the jackknife subsamples, any time point that met the (1) amplitude criteria of a value greater than or equal to the baseline value and (2) the temporal criteria of an averaged value in 10 consecutive 50 ms windows greater than or equal to the baseline value was deemed significant. The first significant time point in each of the jackknife subsamples (n = 15) was used to compute the average latency and 95% confidence interval (one-sided t test). For the latency difference calculations between MEG-model correlation time series, a two-sided paired t test was used. The variance in the standard error calculation was corrected by a factor of (n– 1) [2] to account for the dependence between the jackknifed subsamples. In the latency difference calculations, multiple comparisons were corrected (n = 12 comparisons) using Bonferroni–Holm correction [100] at q = 0.05. The p-values shown here are adjusted for Bonferroni–Holm multiple comparison correction, but the 95% confidence intervals are not adjusted to retain transparency of the original computation and avoid misinterpretations from the step-wise corrections employed by the Bonferroni–Holm method. (XLSX)