Fig. 7: Controlling for locomotion intensity differences between task contexts.

a Example histograms of running speed at three time points during the trial (thin lines) and summary histogram over the time points (thick line). Histogram shows trials from the attend visual context. b Task-context specific running speed histograms. Histograms show summary histograms. Running speed distributions in the two contexts are matched by subsampling trials (shading). c Schematics illustrating run speed matching in the population activity space. Trial-by-trial population responses (circles) on a subspace spanned by the context DV and a direction perpendicular to that. After subsampling trials (filled circles) the distribution of running speed is equal in the two contexts (filled circles). d Mean context decoder accuracy of 10 locomotion matched subsamplings (magenta) and randomized control with same trial numbers (black). Shading shows 2 s.e.m. Gray dashed line indicates shuffled baseline. e Difference between run speed-matched and control decoder accuracies in a population of mice. Whisker plots indicate distribution over time (n_T = 596 timepoints, whiskers between 5–95%, box between 25–75%, notch: median, 95% CI, flyers include all points). f Matched accuracies at single time points along the trial (colored lines), with run speed shifted relative to neural activity (horizontal axis), example animal. g Accuracy of context decoding using all trials (magenta) and correct ‘no-go’ congruent trials only (red) at each timepoint, mean of n = 8 mice (line) and s.e.m. over mice + mean s.e.m.-s from cross-validation variance (bands). h Reconstruction from motion principal components. 6 subsequent frames (left to right): Original capture (top), smoothed difference from a moving average background (upper middle), reconstruction of difference from 40 PCs (lower middle), cumulative sum of the difference reconstruction (bottom). Absolute motion, i.e. mean of absolute values of 40 PCs, and s.e.m (bottom graph). i Accuracy of context decoding using all trials (magenta) and stationary only trials (purple) at each timepoint, s.e.m. over 10-fold cross validation (bands). j Mean context decoder accuracy over the trial time course of stationary trials at different absolute movement thresholds (horizontal axis) and proportion of timepoints allowed to exceed the threshold (colored lines). Mean s.e.m. over the time course (bands). k Mean accuracy over the time course from context-equalized number of trials at various absolute motion level intervals (line), and s.e.m (band).