| Popis: |
Normative modeling frameworks such as Bayesian inference and reward-based learning provide valuable insights into the fundamental principles of adaptive behavior. However, their ability to describe realistic animal behavior is limited by the typically small number of fitted parameters, leading to a cycle of handcrafted adjustments and model comparisons that are prone to research subjectivity. Here, we present a novel modeling approach leveraging recurrent neural networks to automatically discover the cognitive algorithms governing animal decision-making. We show that neural networks with only one or two units can predict choices of individual animals more accurately than classical cognitive models, and as accurately as larger neural networks, in three well-studied reward learning tasks. We then interpret the trained networks using dynamical systems concepts such as state-space and fixed-point attractors, leading to a unified comparison of different cognitive models and a detailed characterization of the cognitive mechanisms underlying the animal’s choices. Our approach also estimates behavior dimensionality and provides insights into the algorithms emerging in meta-reinforcement learning agents. Overall, we present a systematic approach for discovering interpretable cognitive strategies in decision-making, offering insights into neural mechanisms and a foundation for examining both healthy and dysfunctional cognition. |
