Fig. 5.

Evaluating the linearizing encoding models’ prediction accuracy through a pairwise decoding analysis. (A) At each time point (t) we trained an SVM to classify between two BioTest EEG data image conditions (using the channels vectors) and tested it on the two corresponding linearizing SynTest EEG data image conditions. We repeated this procedure across all image condition pairs, and then averaged the decoding accuracies across pairs. This resulted in one decoding score for each time point (red portion in the decoding results toy graph). (B) Pairwise decoding results averaged across participants. The linear classifiers trained on the BioTest data significantly decode the linearizing SynTest data from 60 ms after stimulus onset until the end of the EEG epoch (P < 0.05, one-sample one-sided t-test, Bonferroni-corrected), with peaks at 100–110 ms. (C) Individual participants’ results. Error margins, asterisks, gray area and black dashed lines as in Fig. 4.