Many reinforcement learning (RL) tasks provide the agent with high-dimensional observations that can be simplified into low-dimensional continuous states. To formalize this process, we introduce the concept of a DeepMDP, a parameterized latent space model that is trained via the minimization of two tractable losses: prediction of rewards and prediction of the distribution over next latent states. We show that the optimization of these objectives guarantees (1) the quality of the latent space as a representation of the state space and (2) the quality of the DeepMDP as a model of the environment. We connect these results to prior work in the bisimulation literature, and explore the use of a variety of metrics. Our theoretical findings are substantiated by the experimental result that a trained DeepMDP recovers the latent structure underlying high-dimensional observations on a synthetic environment. Finally, we show that learning a DeepMDP as an auxiliary task in the Atari 2600 domain leads to large performance improvements over model-free RL.
@article{arxiv.1906.02736,
title = {DeepMDP: Learning Continuous Latent Space Models for Representation Learning},
author = {Carles Gelada and Saurabh Kumar and Jacob Buckman and Ofir Nachum and Marc G. Bellemare},
journal= {arXiv preprint arXiv:1906.02736},
year = {2019}
}