Fig. 5: Ensembles encode stimuli more accurately than individual neurons.

a Example responses to four orientations for a neuron and an ensemble tuned to the same preferred orientation (90°). For clarity, 0° trials are mirrored at 180°. b Activation probability, representing the trial-to-trial variability of a neuron and an ensemble activation at their preferred orientation (θ_pref). Black dashed line indicates maximum activation probability of individual neurons versus ensembles. Two-sided Wilcoxon test, p = 9 × 10^–4. c Duration of continuous evoked activity for individual neurons and ensembles after the θ_pref onset. Two-sided Wilcoxon test, p = 5 × 10^–4. d Example responses across all trials for onsemble, offsemble, ensemble^†, and the nonparticipant-tuned population, as in a. Onsemble, offsemble, and nonparticipant-tuned population responses are presented as the fraction of active neurons. Ensemble^† responses are presented as the fraction of ensemble neurons participating in the response (including activated onsemble neurons and inactivated offsemble neurons). e Orientation selectivity. Ensembles showed higher orientation selectivity than other groups (pairwise two-sided Wilcoxon test, p = 0.007 or p = 5 × 10^–4). f Orientation tuning curves fitted for individual neurons, ensembles, onsembles, offsembles, ensembles^†, and the nonparticipant-tuned population. Tuning curves are centered on θ_pref. g Tuning widths, extracted from f. Ensembles showed lower tuning width than the other groups (pairwise two-sided Wilcoxon test, p = 0.007 or p = 5 × 10^–4). Lines and shaded regions represent the mean ± SEM across n = 12 mice. Each datapoint represents the average within each mouse across n = 12 mice. The center of boxplots represents the median, the bounds of the boxes correspond to the first and third quartiles, and the whiskers extend to the minimum and maximum datapoint values. NS not significant, **p < 0.01, and ***p < 0.001. Source data are provided as a Source Data file.