Which Mutual-Information Representation Learning Objectives are Sufficient for Control?

Abstract

Mutual information maximization provides an appealing formalism for learning representations of data. In the context of reinforcement learning (RL), such representations can accelerate learning by discarding irrelevant and redundant information, while retaining the information necessary for control. Much of the prior work on these methods has addressed the practical difficulties of estimating mutual information from samples of high-dimensional observations, while comparatively less is understood about which mutual information objectives yield representations that are sufficient for RL from a theoretical perspective. In this paper, we formalize the sufficiency of a state representation for learning and representing the optimal policy, and study several popular mutual-information based objectives through this lens. Surprisingly, we find that two of these objectives can yield insufficient representations given mild and common assumptions on the structure of the MDP. We corroborate our theoretical results with empirical experiments on a simulated game environment with visual observations.

18 pages, 11 figures

References in corpus (11)

• Learning Representations by Maximizing Mutual Information Across Views
• Variational Intrinsic Control
• Near Optimal Behavior via Approximate State Abstraction
• Loss is its own Reward: Self-Supervision for Reinforcement Learning
• Contrastive Learning of Structured World Models
• DeepMDP: Learning Continuous Latent Space Models for Representation Learning
• Hyperbolic Discounting and Learning over Multiple Horizons
• Independently Controllable Factors
• Is a Good Representation Sufficient for Sample Efficient Reinforcement Learning?
• Comments on the Du-Kakade-Wang-Yang Lower Bounds
• Predictive Coding for Locally-Linear Control