Fig. 3: Illustrations of inverted encoding model (IEM) and procedure.

a First, hypothetical orientation-tuning functions were made for the eight orientation channels, respectively. b These hypothetical orientation-tuning functions were then fed into the orientation responses in each trial of the SO condition. c The EEG signals and orientation responses of the SO condition were used to train the weight matrix, which is termed the SO decoder. d The SO decoder was combined with the EEG signals of the SO and VO conditions to predict the orientation responses of each trial. The SO decoder's prediction of the orientation responses of the SO condition was termed the SO–SO prediction. SO–VO prediction refers to the SO decoder's prediction of the orientation responses of the VO condition. e SO–SO and SO–VO predictions showed that the reconstructed orientation-tuning functions were sharply modulated after the stimulus onset. Top, SO–SO prediction. Bottom, SO–VO prediction. f The orientation-tuning curves were summarized by calculating the slopes of curves. Specifically, we first reversed the sign of orientations. We then calculate the slopes by fitting the linear regression with intercept. g–h Orientation decoding results of SO–SO prediction during Experiment 1 (h) and Experiment 2 (g). The orientation sensitivity were sharply increased after the stimuli onset. i–j Orientation decoding results of SO–VO predictions during Experiment 1 (i) and Experiment 2 (j). The significant modulations showed the reconstruction of mean orientations. The x-axis shows the time after stimulus onset. The y-axis shows the orientation sensitivity. The gray rectangle from 0 to 0.1 s after the stimulus onset shows the period of the presentation of the stimuli. Blue shaded areas represent ± 1 bootstrapped standard error. Blue bars above the figures (g–j) show significant modulation over time, determined by a one-sided Monte Carlo randomization test, p < 0.05. SO same orientation, VO varied orientation.