REPAINT: Knowledge Transfer in Deep Reinforcement Learning
This work addresses the problem of accelerating learning for complex reinforcement learning tasks, particularly when source and target tasks have low similarity, which is a common challenge for practitioners in RL.
This paper introduces REPAINT, a knowledge transfer algorithm for deep reinforcement learning that combines representation transfer from a pre-trained teacher policy with advantage-based experience selection for off-policy learning. It significantly reduces training time across various task similarities, outperforming baselines, especially when source tasks are dissimilar or sub-tasks.
Accelerating learning processes for complex tasks by leveraging previously learned tasks has been one of the most challenging problems in reinforcement learning, especially when the similarity between source and target tasks is low. This work proposes REPresentation And INstance Transfer (REPAINT) algorithm for knowledge transfer in deep reinforcement learning. REPAINT not only transfers the representation of a pre-trained teacher policy in the on-policy learning, but also uses an advantage-based experience selection approach to transfer useful samples collected following the teacher policy in the off-policy learning. Our experimental results on several benchmark tasks show that REPAINT significantly reduces the total training time in generic cases of task similarity. In particular, when the source tasks are dissimilar to, or sub-tasks of, the target tasks, REPAINT outperforms other baselines in both training-time reduction and asymptotic performance of return scores.