The abstract is read at the publisher; the summary is JClub's.
A hierarchical recurrent neural network under the Free Energy Principle reproduces WCST stability and flexibility by keeping category estimates after correct feedback and switching after incorrect feedback.
- Why it matters: Existing WCST models often assume category variables in advance and treat correct feedback as inherently desirable, leaving the joint learning-and-goal mechanism unclear.
- What they did: The model performs goal-directed inference and planning: it infers the environment from observations and then plans card selection, minimizing prediction error for inference and for planning toward correct feedback.
- The result: During learning, latent states form category-related clusters, and in testing the model keeps after correct feedback and switches after incorrect feedback; removing goal-directed planning degrades both.
The findingWhy it mattersWhat they didThe result
AI summary of the abstract · may contain errors · not medical advice · read the paper.