Fig. 3: Population codes and their relation to coupling filter remapping and behavior.
A Illustration of task-unaligned latent neural dynamics. Trials were categorized according to their target location (top: 9 bins, from near-to-far and from leftward-to-rightward) and density condition (bird’s eye view, first row). Second, third, and fourth row respectively show the averaged latent trajectory in 2D for MSTd (green), 7a (blue), and dlPFC (red), both during high (colored) and low (black) density conditions, and according to target location. B Decoding of the distance to target based on neural population dynamics within condition (high-density/high-density, low-density/low-density) or across (high-density/low-density). Dots are individual sessions, dashed black line is cross-validated R2 = 0. C Five example trials decoding distance to target. D Decoding of the distance to target based on populations where signal correlations were left intact but noise correlations were destroyed. E Relation between stability of the fine-grain neural co-fluctuations and ability of the decoder to generalize. F Relation between stability of the fine-grain neural co-fluctuations and session-by-session changes in behavioral performance with optic flow density. Source data are provided as a Source data file.
