Fig. 4: Decision-related neural representations in the reverse task.

a, d Neural trajectories projected onto the principal component of the decision-related subspace for Task A and Reverse-A, shown for a representative recording session. b, e The similarity between the representation of the principal component of the decision subspace manifold in A/B/Revisit-A tasks and in the Reverse-A task. Similarity was quantified using Pearson correlation. Data are presented as mean ± SD. Error bars indicate standard deviation across three recording sessions. Each dot represents the similarity for one direction (up or down) manifold within a recording session, yielding six data points in total. (n = 3 sessions per monkey; two-tailed t-test on Pearson correlation coefficients, Monkey ZZ: P = 0.03, g = 1.47; Monkey AB: P = 0.002, g = 2.36). c, f A linear relationship exists between the similarity of the representation of the decision subspace manifold and the number of trials required to learn the tasks. Each small, filled circle represents data from one recording session for a specific task condition (color-coded). Large hollow circles show condition averages. Cross-shaped error bars indicate standard deviations in both axes over three sessions. The coefficient of determination (R²) and p-value of the regression model compared to the null model are indicated on the graph. (n = 3 sessions per monkey; linear regression, Monkey ZZ: P = 0.04, R² =0.47; Monkey AB: P = 0.04, R² = 0.49). *, P < 0.05; **, P < 0.01; ***, P < 0.001.