Mean cosine similarity for pairs of axes in neural activity space (Chrimson n = 17 sessions, ArchT n = 16 sessions, control n = 17 sessions, error bars indicate SEM). From left to right, absolute cosine similarity for: LDA axis separating VIP photoactivation vs. no photoactivation trials and LDA axis separating attend vs. ignore trials; pairs of random axes extracted based on the covariance of the original neural data (10,000 samples); two axes separating the rewarded vs. unrewarded visual stimuli where each axis was found using half of the data (mean of 50 shuffled repeats). Significance tests were against the corresponding random axes median using Wilcoxon signed-rank test,*p < 0.01,***p < 0.001. From left to right p values for all tests: 0.009, 0.379, 0.653, 6.43 x 10−4, 2.93 x 10−4, 0.002.
