mann-whitney u test matlab ranksum command  (MathWorks Inc)

Journal: Nature human behaviour

Article Title: Multiplexing working memory and time: encoding retrospective and prospective information in neural trajectories

doi: 10.1038/s41562-023-01592-y

Figure Lengend Snippet: a, Confusion matrices of SVM decoding of Cue and time bins (100 ms) for a sample RNN trained on each task. Note that in this example, the ISA network often confuses time bins from approximately 0.5-1 sec during the short (Cue=A) and long (Cue=B) delays as reflected in the off-diagonal bands. b, Performance (left) and MSE (right) of the decoders across all RNNs. All pairwise comparisons for both Performance and MSE were significant at p<10−5 (Wilcoxon rank sum tests). c, Correlation between unit activity for sample RNNs during the short and long delays in the WM (left), T+WM (middle), and ISA (right) tasks. d. For the WM and T+WM tasks there was little or no average correlation across all units within each RNN (n=17 in each group). In the ISA task average correlation between unit activity in the short and long delay was high and significantly above the WM and T+WM tasks (p<10−4, Wilcoxon rank sum tests). Additionally, only the ISA correlation was significantly different than 0 (p<0.0005, sign rank tests).

Article Snippet: Comparison across RNN tasks relied on the nonparametric Wilcoxon rank sum test ( .ranksum command in Matlab).

Techniques: Activity Assay