Figure 6.

Cursor identity was more decodable from neural activity during task performance

(A) Mean firing rate to different cursor positions depended on whether the preceding cursor position was sweeping toward (red) or away (black) from the direction of the target during task (t test, Bonferroni corrected, ^∗p < 0.05 and ^∗∗p < 0.01; error bars indicate SEM; n = 131 units).

(B) Cursor sweep direction had little effect on firing rates when animals were passively viewing playback (t test, Bonferroni corrected, ^∗∗p < 0.01; error bars indicate SEM; n = 128 units.)

(C) Classification accuracy for cursor identity, from population responses trained on real (blue) and shuffled data (gray), for neural responses during task performance (left panel) and passive playback (right panel). Shaded regions denote 95% confidence interval (n = 7 mice). During task performance, the classifier could infer the upcoming cursor position even before presentation (chance level at 12.5%), and rose higher after presentation, suggesting that neural responses encode expectation. This was not true of the passive playback condition, in which case the classifier only performed above chance during the cursor presentation period.

(D) Classification accuracy for cursor direction, from population responses trained on real (blue) and shuffled labels data (gray), for neural responses during task performance (left panel) and passive playback (right panel). Shaded regions denote 95% confidence interval (n = 7 mice). A trained classifier could not perform above chance (50%) in predicting, from firing alone, the direction that the visual cursor was moving, in either the task or playback conditions.