Fig. 7: Studying efficient allocation of neural resources with non-monotonic stimulus–reward mappings.
a,b, The prior distribution of sensory stimuli in the environment monotonically decreases with sensory stimuli (black) and is the same in all scenarios. The stimulus–reward mapping function in scenario 2 monotonically increases following a linear relationship (red, a) and in scenario 3 is non-monotonic with the highest reward delivered at s = 0.5 (green, b). Scenario 1 corresponds to the accuracy maximization context—that is, any correct decision yields the same amount of reward. c, Optimal solutions of the resource allocation problem for scenario 1 (blue), scenario 2 (red) and scenario 3 (green). d, Percentage reward lost in scenario 3 assuming that the agent uses the optimal resource allocations from Kacc (MaxAcc, blue), Krew (MaxRew, red) or environments relative to the optimal solution in this non-monotonic stimulus–reward mapping environment (Non-Mon, green).
