Box Plot from Scientific Research

The difference between predicted and actual state values (the mean squared error) to rewards near the center point (reward size of 5) of the training distribution, for each projection-defined agent. DLS- and VLS-projecting populations made significantly worse value estimates than DMS- or NAc-projecting populations. DMS- and NAc-projecting populations were more accurate than the unified population, whereas VLS made significantly larger errors.*p< 0.0001 Kruskal-Wallis one-way ANOVA on ranks with Tukeys post-hoc test.

DOI