Minimal model of strategy switching in the plus-maze navigation task

Abstract. Prefrontal cortex (PFC) has been implicated in the ability to switch behavioral strategies in response to changes in reward contin-gencies. A recent experimental study has shown that separate subpop-ulations of neurons in the prefrontal cortex were activated when rats switched between allocentric place strategies and egocentric response strategies in the plus maze. In this paper we propose a simple neural-network model of strategy switching, in which the learning of the two strategies as well as learning to select between those strategies is gov-erned by the same temporal-difference (TD) learning algorithm. We show that the model reproduces the experimental data on both behavioral and neural levels. On the basis of our results we derive testable prediction concerning a spatial dynamics of the phasic dopamine signal in the PFC, which is thought to encode reward-prediction error in the TD-learning theory.