Free Energy Principle
Overview
The Free Energy Principle (FEP) describes how agents can maintain workable models of their surroundings by reducing variational free energy, a quantity related to prediction error and uncertainty. Active inference extends this idea to action: agents use expected free energy to balance achieving preferred outcomes with gathering useful information. Methods include generative models and message passing for belief-state planning, tested in settings such as grid-world POMDPs, visual navigation, and robotic control. A central challenge is defining and calibrating the balance between goal-directed and information-seeking behavior, while showing that these principles scale reliably beyond controlled benchmarks.
Latest papers
20 papers1 paper in the last four weeks, against 1 the four weeks before. 0.0% of all new papers.