Uncertainty-Based Offline Reinforcement Learning with Diversified Q-Ensemble

Offline reinforcement learning (offline RL), which aims to find an optimal policy from a previously collected static dataset, bears algorithmic difficulties due to function approximation errors from out-of-distribution (OOD) data points. To this end, offline RL algorithms adopt either a constraint or a penalty term that explicitly guides the policy to stay close to the given dataset. However, prior methods typically require accurate estimation of the behavior policy or sampling from OOD data points, which themselves can be a non-trivial problem. Moreover, these methods under-utilize the generalization ability of deep neural networks and often fall into suboptimal solutions too close to the given dataset. In this work, we propose an uncertainty-based offline RL method that takes into account the confidence of the Q-value prediction and does not require any estimation or sampling of the data distribution. We show that the clipped Q-learning, a technique widely used in online RL, can be leveraged to successfully penalize OOD data points with high prediction uncertainties. Surprisingly, we find that it is possible to substantially outperform existing offline RL methods on various tasks by simply increasing the number of Q-networks along with the clipped Q-learning. Based on this observation, we propose an ensemble-diversified actor-critic algorithm that reduces the number of required ensemble networks down to a tenth compared to the naive ensemble while achieving state-of-the-art performance on most of the D4RL benchmarks considered.

[1]  Sergey Levine,et al.  D4RL: Datasets for Deep Data-Driven Reinforcement Learning , 2020, ArXiv.

[2]  Martha White,et al.  Selective Dyna-style Planning Under Limited Model Capacity , 2020, ICML.

[3]  Richard Y. Chen,et al.  UCB EXPLORATION VIA Q-ENSEMBLES , 2018 .

[4]  Thorsten Joachims,et al.  MOReL : Model-Based Offline Reinforcement Learning , 2020, NeurIPS.

[5]  David Silver,et al.  Deep Reinforcement Learning with Double Q-Learning , 2015, AAAI.

[6]  Nitish Srivastava,et al.  Uncertainty Weighted Actor-Critic for Offline Reinforcement Learning , 2021, ICML.

[7]  Shane Legg,et al.  Human-level control through deep reinforcement learning , 2015, Nature.

[8]  Jasper Snoek,et al.  Practical Bayesian Optimization of Machine Learning Algorithms , 2012, NIPS.

[9]  Doina Precup,et al.  Off-Policy Deep Reinforcement Learning without Exploration , 2018, ICML.

[10]  S. Levine,et al.  Conservative Q-Learning for Offline Reinforcement Learning , 2020, NeurIPS.

[11]  William R. Clements,et al.  Estimating Risk and Uncertainty in Deep Reinforcement Learning , 2019, ArXiv.

[12]  Csaba Szepesvári,et al.  Exploration-exploitation tradeoff using variance estimates in multi-armed bandits , 2009, Theor. Comput. Sci..

[13]  Sergey Levine,et al.  Stabilizing Off-Policy Q-Learning via Bootstrapping Error Reduction , 2019, NeurIPS.

[14]  J. Royston.  Expected Normal Order Statistics (Exact and Approximate) , 1982 .

[15]  Robert Loftin,et al.  Better Exploration with Optimistic Actor-Critic , 2019, NeurIPS.

[16]  Yifan Wu,et al.  Behavior Regularized Offline Reinforcement Learning , 2019, ArXiv.

[17]  Marc G. Bellemare,et al.  The Arcade Learning Environment: An Evaluation Platform for General Agents , 2012, J. Artif. Intell. Res..

[18]  Martha White,et al.  Maxmin Q-learning: Controlling the Estimation Bias of Q-learning , 2020, ICLR.

[19]  David S. Watkins,et al.  Understanding the $QR$ Algorithm , 1982 .

[20]  Mohammad Norouzi,et al.  An Optimistic Perspective on Offline Reinforcement Learning , 2020, ICML.

[21]  Benjamin Van Roy,et al.  Deep Exploration via Bootstrapped DQN , 2016, NIPS.

[22]  Yuan Qi,et al.  Generative Adversarial User Model for Reinforcement Learning Based Recommendation System , 2018, ICML.

[23]  Pieter Abbeel,et al.  SUNRISE: A Simple Unified Framework for Ensemble Learning in Deep Reinforcement Learning , 2021, ICML.

[24]  Herke van Hoof,et al.  Addressing Function Approximation Error in Actor-Critic Methods , 2018, ICML.

[25]  Sergey Levine,et al.  Soft Actor-Critic: Off-Policy Maximum Entropy Deep Reinforcement Learning with a Stochastic Actor , 2018, ICML.

[26]  Albin Cassirer,et al.  Randomized Prior Functions for Deep Reinforcement Learning , 2018, NeurIPS.

[27]  Lantao Yu,et al.  MOPO: Model-based Offline Policy Optimization , 2020, NeurIPS.

[28]  S. Levine,et al.  Offline Reinforcement Learning: Tutorial, Review, and Perspectives on Open Problems , 2020, ArXiv.