Observational Overfitting in Reinforcement Learning

Abstract

A major component of overfitting in model-free reinforcement learning (RL) involves the case where the agent may mistakenly correlate reward with certain spurious features from the observations generated by the Markov Decision Process (MDP). We provide a general framework for analyzing this scenario, which we use to design multiple synthetic benchmarks from only modifying the observation space of an MDP. When an agent overfits to different observation spaces even if the underlying MDP dynamics is fixed, we term this observational overfitting. Our experiments expose intriguing properties especially with regards to implicit regularization, and also corroborate results from previous works in RL generalization and supervised learning (SL).

Published as a conference paper in ICLR 2020

References in corpus (6)

• StarCraft II: A New Challenge for Reinforcement Learning
• Robust Adversarial Reinforcement Learning
• Implicit Regularization in Deep Learning
• Generalization in Reinforcement Learning with Selective Noise Injection and Information Bottleneck
• Generalization in Deep Networks: The Role of Distance from Initialization
• On the Generalization Gap in Reparameterizable Reinforcement Learning