Fig. 5: Estimation task performance.
a, Changes in estimation variability for Kacc (blue) and Krew (red) stimulus–reward association contexts. Estimation variability before training is shown in the dotted lines and open triangles, while post-training variability is shown in the solid lines and filled circles. The results from Krew training show the interaction between oblique (decreased variance) and cardinal angles (increased variance) predicted by the implementation of a fitness-maximizing coding scheme for linear stimulus–reward mappings. In contrast, Kacc training produces no such interaction. b, Changes in estimation variability for the Kacc and Krew contexts in trained (top row) and untrained locations (bottom row) in experiment 2 indicate retinotopically specific training effects. The Krew training leads to an interaction between oblique and cardinal variance changes in trained locations only. Training Kacc does not lead to this interaction in either location.
