Suppressing Overestimation in Q-Learning through Adversarial Behaviors

TL;DR

This paper introduces DAQ, a novel Q-learning algorithm with a dummy adversarial player that effectively reduces overestimation bias by framing learning as a two-player game, improving performance in benchmark tasks.

Contribution

The paper proposes a new adversarial Q-learning framework that unifies and enhances existing methods to suppress overestimation bias in Q-learning algorithms.

Findings

DAQ effectively reduces overestimation bias.

Finite-time convergence of DAQ is theoretically analyzed.

Empirical results show improved performance on benchmarks.

Abstract

The goal of this paper is to propose a new Q-learning algorithm with a dummy adversarial player, which is called dummy adversarial Q-learning (DAQ), that can effectively regulate the overestimation bias in standard Q-learning. With the dummy player, the learning can be formulated as a two-player zero-sum game. The proposed DAQ unifies several Q-learning variations to control overestimation biases, such as maxmin Q-learning and minmax Q-learning (proposed in this paper) in a single framework. The proposed DAQ is a simple but effective way to suppress the overestimation bias thourgh dummy adversarial behaviors and can be easily applied to off-the-shelf reinforcement learning algorithms to improve the performances. A finite-time convergence of DAQ is analyzed from an integrated perspective by adapting an adversarial Q-learning. The performance of the suggested DAQ is empirically demonstrated under various benchmark environments.