Predictive Coding Enhances Meta-RL To Achieve Interpretable Bayes-Optimal Belief Representation Under Partial Observability

Will Dabney (Google DeepMind) · Po-Chen Kuo (University of Washington) · Han Hou (Allen Institute for Neural Dynamics) · Edgar Walker (University of Washington)
active information seekingadaptabilityauxiliary predictive objectivesbayes-optimal policiesbayesian inferencebelief stateseffective representation learninggeneralization capacityinterpretable representationsmeta-reinforcement learningpartial observabilitypredictive codingrepresentation learningrepresentational inefficiencyself-supervised learning

Learning a compact representation of history is critical for planning and generalization in partially observable environments. While meta-reinforcement learning (RL) agents can attain near Bayes-optimal policies, they often fail to learn the compact, interpretable Bayes-optimal belief states. This representational inefficiency potentially limits the agent's adaptability and generalization capacity. Inspired by predictive coding in neuroscience