agentsfactbullishActive Inference can be formulated as a convex Markov decision process for closed-loop control policiesMachine Learning (Statistics)28 Jul 2026http://arxiv.org/abs/2607.20152v1