← Back to papers
2018 ICML ICML 2018

Self-Imitation Learning

Abstract

This paper proposes Self-Imitation Learning (SIL), a simple off-policy actor-critic algorithm that learns to reproduce the agent’s past good decisions. This algorithm is designed to verify our hypothesis that exploiting past good experiences can indirectly drive deep exploration. Our empirical results show that SIL significantly improves advantage actor-critic (A2C) on several hard exploration Atari games and is competitive to the state-of-the-art count-based exploration methods. We also show that SIL improves proximal policy optimization (PPO) on MuJoCo tasks.

πŸŒ‰ Interdisciplinary Bridge β€” Artificial Intelligence and Reinforcement Learning
πŸ“ˆ Trend Setter β€” Few-Shot Learning
🧭 Keyword Pioneer β€” self-imitation learning
🐝 Cross-Pollinator β€” Artificial Intelligence, Machine Learning, Reinforcement Learning
🐣 Hot Topic Early Bird β€” proximal policy optimization

Authors

Junhyuk Oh , Yijie Guo , Satinder Singh , Honglak Lee

Topics

Artificial Intelligence > Learning Paradigms > Few-Shot Learning Reinforcement Learning > Methods > Deep RL Reinforcement Learning > Methods > Offline RL Reinforcement Learning > Applications > Game AI Reinforcement Learning > Applications > Robotics

Keywords

proximal policy optimization self-imitation learning deep exploration off-policy actor-critic past good decision advantage actor-critic hard exploration game
Download PDF

Related papers

Rectify Heterogeneous Models with Semantic Mapping 2018
Bayesian Optimization of Combinatorial Structures 2018
The Well-Tempered Lasso 2018
Approximation Algorithms for Cascading Prediction Models 2018
Classification from Pairwise Similarity and Unlabeled Data 2018