Deep Active Inference for Partially Observable MDPs

Abstract

Deep active inference has been proposed as a scalable approach to perception and action that deals with large policy and state spaces. However, current models are limited to fully observable domains. In this paper, we describe a deep active inference model that can learn successful policies directly from high-dimensional sensory inputs. The deep learning architecture optimizes a variant of the expected free energy and encodes the continuous state representation by means of a variational autoencoder. We show, in the OpenAI benchmark, that our approach has comparable or better performance than deep Q-learning, a state-of-the-art deep reinforcement learning algorithm.

1st International Workshop on Active inference, European Conference on Machine Learning (ECML/PCKDD 2020)

References in corpus (3)

• A Brief Survey of Deep Reinforcement Learning
• End-to-End Pixel-Based Deep Active Inference for Body Perception and Action
• A deep active inference model of the rubber-hand illusion

Cited by in corpus (3)

• Geometric Methods for Sampling, Optimisation, Inference and Adaptive Agents
• Active Inference for Stochastic Control
• Sensorimotor Visual Perception on Embodied System Using Free Energy Principle