Extended Data Fig. 2: Performance simulations with probabilistic choice outcomes.
From: Asymmetric reinforcement learning facilitates human inference of transitive relations
Same conventions as in Fig. 2. Left, full feedback (for all pairs; cf. Figure 2b). Right, partial feedback (only for non-neighbouring pairs; cf. Figure 2e). Optimal learning under partial feedback is characterized by asymmetric updating (α+ ≠ α−), just as was observed with deterministic outcomes (cf. Figure 2e).
