Tactical Optimism and Pessimism for Deep Reinforcement Learning
arXiv:2102.03765
Abstract
In recent years, deep off-policy actor-critic algorithms have become a dominant approach to reinforcement learning for continuous control. One of the primary drivers of this improved performance is the use of pessimistic value updates to address function approximation errors, which previously led to disappointing performance. However, a direct consequence of pessimism is reduced exploration, running counter to theoretical support for the efficacy of optimism in the face of uncertainty. So which approach is best? In this work, we show that the most effective degree of optimism can vary both across tasks and over the course of learning. Inspired by this insight, we introduce a novel deep actor-critic framework, Tactical Optimistic and Pessimistic (TOP) estimation, which switches between optimistic and pessimistic value learning online. This is achieved by formulating the selection as a multi-arm bandit problem. We show in a series of continuous control tasks that TOP outperforms existing methods which rely on a fixed degree of optimism, setting a new state of the art in challenging pixel-based environments. Since our changes are simple to implement, we believe these insights can easily be incorporated into a multitude of off-policy algorithms.
References in corpus (19)
- Continuous control with deep reinforcement learning
- Soft Actor-Critic: Off-Policy Maximum Entropy Deep Reinforcement Learning with a Stochastic Actor
- DeepMind Control Suite
- Learning Latent Dynamics for Planning from Pixels
- Implicit Quantile Networks for Distributional Reinforcement Learning
- Agent57: Outperforming the Atari Human Benchmark
- Provably Efficient Reinforcement Learning with Linear Function Approximation
- Distributional Reinforcement Learning with Quantile Regression
- REGAL: A Regularization based Algorithm for Reinforcement Learning in Weakly Communicating MDPs
- Effective Diversity in Population Based Reinforcement Learning
- Better Exploration with Optimistic Actor-Critic
- Model Selection in Contextual Stochastic Bandit Problems
- Evolving Reinforcement Learning Algorithms
- Regret Bound Balancing and Elimination for Model Selection in Bandits and RL
- Near-optimal Optimistic Reinforcement Learning using Empirical Bernstein Inequalities
- Adapting Behaviour for Learning Progress
- Ready Policy One: World Building Through Active Learning
- Learning to Score Behaviors for Guided Policy Optimization
- A Scheme for Dynamic Risk-Sensitive Sequential Decision Making