Fig. 3: I Computational modelling of prediction error trajectories using an ideal observer model.
From: Predictive learning shapes the representational geometry of the human brain
a analysis framework for fitting prediction error trajectories from the model to each parcel of the brain. Left: For each time point relative to a tone presentation, the brain data is structured as a matrix with parcels and tones. Right: prediction error trajectory extracted from the ideal observer model. Centre: illustration of Gaussian Copula Mutual Information (GCMI). b Left: prediction error trajectories extracted from the model for both sequences. Middle: prediction accuracies of the model. Right: weight matrices of the ideal observer model after training on the full sequences. c top: Time-course of prediction error encoding (GCMI) for both sequences. Shaded areas indicate SEM, and horizontal lines indicate statistical significance (p < 0.05). Significance was determined by cluster-based paired two-tailed permutation tests. Bottom: cortical distribution of prediction error encoding in 4 time windows. d correlation analysis between the representational shift and the prediction error encoding for the brain regions indicated on the top. Dots represent individual participants. Shaded areas indicate 95% confidence intervals. The dotted lines indicate non-significant regression fits. Solid lines indicate significant regression fits. Significance (p < 0.05) was determined using the Pearson correlation coefficient and Bonferroni correction. Source data are provided as a Source Data file.
