Insular cortex encodes task alignment

Animals can learn complex behaviors. Animal behavior in the lab has traditionally been studied via summary statistics such as trial-based success rates. However, animal behavior is much more fine-grained: a trial in an experiment often consists of multiple actions, and more than one strategy can lead to a successful completion of a trial. To understand how the brain controls behavior, a fine-grained yet compact description of behavior is necessary. We describe here an approach for estimating the strategies animals use for a large family of tasks from first principles. Using reinforcement learning with informational constraints on policies, we compute a rich set of candidate policies with a small number of meaningful parameters, and match observed behavior to these policies. In a sample rat task, our approach revealed ongoing learning for more than 100 days after the saturation of success rates. Moreover, we showed that many neurons in the insular cortex of rats track the instantaneous task engagement of the rats with a resolution of a few minutes. Due to its generic formulation in reinforcement learning terminology, our work is directly applicable to the majority of animal tasks in use today.