A neural network model of encoding rules in the prefrontal cortex

In order to investigate encoding of rules in the prefrontal (PF) cortex, we simulated the physiological results of Wallis et al. (2001). They explored its neural basis by recording from single neurons in the PF cortex of monkeys trained to use two rules. As a result, they found neurons selective for learned rules regardless of samples and cues. However, the functional role of the rule-selective neurons has not yet been elucidated. To investigate how the brain may implement rule-guided behaviour, a fully recurrent neural network model was optimized to perform a rule-guided delayed matching-to-sample task. In the model´s hidden layer rule-selective units, object-selective units and complex units were found and an examination of connection weights showed that rule-selective neurons maintain encoded rule information, while complex units contribute to outputs.