Can Agents Learn by Analogy? An Inferable Model for PAC Reinforcement Learning

TL;DR

This paper introduces GIM, a novel model-based reinforcement learning method that infers unknown environment dynamics from known ones using spectral properties, enabling efficient learning by analogy.

Contribution

The paper presents GIM, a new inference-based algorithm that improves efficiency and sample complexity in reinforcement learning by leveraging environment spectral properties.

Findings

GIM outperforms existing algorithms in real-world tasks.

GIM reduces computational complexity independent of environment size.

GIM achieves lower sample complexity under mild conditions.

Abstract

Model-based reinforcement learning algorithms make decisions by building and utilizing a model of the environment. However, none of the existing algorithms attempts to infer the dynamics of any state-action pair from known state-action pairs before meeting it for sufficient times. We propose a new model-based method called Greedy Inference Model (GIM) that infers the unknown dynamics from known dynamics based on the internal spectral properties of the environment. In other words, GIM can "learn by analogy". We further introduce a new exploration strategy which ensures that the agent rapidly and evenly visits unknown state-action pairs. GIM is much more computationally efficient than state-of-the-art model-based algorithms, as the number of dynamic programming operations is independent of the environment size. Lower sample complexity could also be achieved under mild conditions compared against methods without inferring. Experimental results demonstrate the effectiveness and efficiency of GIM in a variety of real-world tasks.