Skip to main content
. 2018 Mar 8;7:e32259. doi: 10.7554/eLife.32259

Figure 8. The single-unit correlates of the weighted linear decoding scheme.

Figure 8.

(a–b) Distributions of unit d’, computed for each monkey. Arrows indicate means. (c) Left, Comparison of behavioral prediction error trajectories for an FLD decoder applied to randomly selected units (replotted from Figures 6g and 7a) versus when the top-ranked units for each population size were selected before cross-validated testing. Dots correspond to population sizes with lowest error. Right, Conversion of behavioral error predictions (MSE) into prediction quality (PQ). Dots indicate PQ for the best sized population and for all units. (d) Left, Comparison of behavioral prediction error trajectories for an SCC decoder applied to randomly selected units (replotted from Figure 7a) versus when the top-ranked units for each population size were selected before cross-validated testing. Dots correspond to population sizes with lowest error. Right, Conversion of MSE into PQ. Black dot indicates PQ for the best sized population for randomly selected units. Red dots indicate PQ for the 400 top-ranked, positive sign (d’>0) units, all positive sign units, and all units.