Prior preference learning from experts: Designing a reward with active inference

Active inference may be defined as Bayesian modeling of a brain with biologically plausible model the agent. Its primary idea relies on free energy principle and prior preference An agent will choose an action that leads to its for future observation. In this paper, we claim active can interpreted using reinforcement learning (RL) algorithms find theoretical connection between them. We extend concept expected (EFE), which is core quantity in inference, EFE treated negative value function. Motivated by connection, propose simple but novel method from experts. This illustrates problem inverse RL approached new perspective inference. Experimental results show possibility EFE-based rewards application problem.