Fig. 2
Identification of a subspace. a Illustration of the subspace where Delay 1 and Delay 2 activity exists as one persistent code. b Delay 1 (plotted using crosses) and Delay 2 (plotted using squares) responses after projection into the top 2 dimensions of the subspace. Points for different target locations are color-coded according to the color scheme shown in the top left. c Heat map showing the cross-temporal population-decoding performance after the population responses were projected onto the subspace. White lines indicate target presentation (0–0.3 s) and distractor presentation (1.3–1.6 s). d The difference in decoding performance when a decoder that was trained on Delay 1 responses was tested on Delay 1 (LP11) or Delay 2 (LP12) responses are shown in the gray box-plot labelled LP11–LP12, and vice versa (gray box-plot labelled LP22–LP21). The equivalent performance differences in the subspace are shown in the black box-plots. e The shift in cluster centers from Delay 1 to Delay 2 (labeled D1–D2) averaged across target locations, from Delay 1 to the distractor presentation period (D1 - Dist), and the intra-delay shifts in both Delays 1 and 2 (intra-delay) are shown in the full space (gray box-plot), and in the subspace (black box-plot). f The cumulative explained variance is plotted as a function of the number of PCs for the full space (plotted in gray) and the subspace (plotted in black). The explained variance denoted the dimensionality of the projections in the full space and the subspace (not to be confused with the 58 dimensions that explained 90% of the variance in normalized z-scored data). This was performed to show that the dimensionality of the subspace was lower than that of the full space. The bounds in the boxplots are defined by the 25th and 75th percentile of the distribution. The red line represents the median and the whiskers represent the 2.5th and 97.5th percentile of the distribution
