Simulation of Wisconsin Card Sorting Task by using Recurrent Neural Network based on free energy principle

Abstract

Wisconsin Card Sorting Test (WCST) is a psychological task to measure cognitive flexibility. Computational modeling of underlying cognitive mechanisms of WCST is important for elucidating flexible cognitive processing. Here, we propose a hierarchical recurrent neural network model for explaining mechanisms of WCST, based on the free-energy principle (FEP). FEP, which explains perception and action as minimization of prediction errors between predicted and real sensory inputs, is expected as an integrated theory of the brain. The main characteristic of our model is to consider free energy of the future. It enables the model to correctly answer WCST via goal-directed action. As a result of the simulation experiment, the proposed model successfully estimated the target category and changed the answer along with the change of the target category. This result shows that the proposed model may provide mechanistic insights into flexible cognitive processing from the viewpoint of FEP.