Fig. 2: Model of large-scale cortical dynamics.

A Model schematics. x represents the activity time series of a cortical parcel and \(u\) represents input from the stimulus. Model parameters include directional interactions between neural units (\(W\)), self-decay that determines autocorrelation (\(D\)), and the stimulus-to-brain relationship (\(\beta\)). Although only two units are visualized for simplicity, the model was fit on the time series of 200 cortical parcels. B Model optimization. The model was trained to minimize the difference between the observed and predicted neural activity patterns of consecutive time steps. Green denotes parameters that are estimated during training, black lines denote observed neural activity pattern at a time step (\({x}_{t}\)), with orange indicating sigmoidal bound from −1 to 1 given the nonlinear transfer function, and blue denotes stimulus embeddings at the corresponding time step (\({u}_{t}\)).