Behavior of networks optimized to perform classical conditioning on a reversal learning task.

(A) Top: Schematic of reversal learning task. In the first phase, CS1 but not CS2 is paired with US, while during reversal the contingencies are reversed. Preference between CS1 and CS2 is compared in the test phase. Bottom: Example MBON and DAN activity during reversal learning. (B) The average difference in reported valence for CS2 vs. CS1. Positive or negative values for positive or negative-valence US, respectively, indicate successful reversal learning. Bars indicate standard deviation across model networks. (EPS)