Self-Imitation Learning via Generalized Lower Bound Q-learning

Abstract

Self-imitation learning motivated by lower-bound Q-learning is a novel and effective approach for off-policy learning. In this work, we propose a n-step lower bound which generalizes the original return-based lower-bound Q-learning, and introduce a new family of self-imitation learning algorithms. To provide a formal motivation for the potential performance gains provided by self-imitation learning, we show that n-step lower bound Q-learning achieves a trade-off between fixed point bias and contraction rate, drawing close connections to the popular uncorrected n-step Q-learning. We finally show that n-step lower bound Q-learning is a more robust alternative to return-based self-imitation learning and uncorrected n-step, over a wide range of continuous control benchmark tasks.

Accepted at NeurIPS (Neural Information Processing Systems) 2020, Vancouver, Canada. Code is available at https://github.com/robintyh1/nstep-sil

References in corpus (8)

• DeepMind Control Suite
• Rainbow: Combining Improvements in Deep Reinforcement Learning
• Distributed Distributional Deterministic Policy Gradients
• Averaged-DQN: Variance Reduction and Stabilization for Deep Reinforcement Learning
• Learning to Play in a Day: Faster Deep Reinforcement Learning by Optimality Tightening
• Learning Self-Imitating Diverse Policies
• Multi-step Off-policy Learning Without Importance Sampling Ratios
• Adaptive Trade-Offs in Off-Policy Learning

Cited by in corpus (4)

• Exploration in Deep Reinforcement Learning: From Single-Agent to Multiagent Domain
• Self-Imitation Advantage Learning
• Hindsight Expectation Maximization for Goal-conditioned Reinforcement Learning
• Qgraph-bounded Q-learning: Stabilizing Model-Free Off-Policy Deep Reinforcement Learning