Active Inference in Discrete State Spaces from First Principles
This work addresses a conceptual issue in theoretical neuroscience and AI by providing a more foundational understanding of active inference, which is incremental in refining existing frameworks.
The paper tackles the problem of clarifying active inference by separating it from the Free Energy Principle, showing that optimizations in discrete state spaces can be formulated as constrained divergence minimization problems solvable by standard mean field methods, with the proposed criterion coinciding with variational free energy for perception and differing by an entropy regularizer for action.
We seek to clarify the concept of active inference by disentangling it from the Free Energy Principle. We show how the optimizations that need to be carried out in order to implement active inference in discrete state spaces can be formulated as constrained divergence minimization problems which can be solved by standard mean field methods that do not appeal to the idea of expected free energy. When it is used to model perception, the perception/action divergence criterion that we propose coincides with variational free energy. When it is used to model action, it differs from an expected free energy functional by an entropy regularizer.