Continual Offline Reinforcement Learning with Decision Transformer: Addressing Catastrophic Forgetting
Decision Transformer can serve as a more suitable offline continuous learner compared to Actor-Critic based algorithms, but faces challenges with catastrophic forgetting. The proposed methods MH-DT and LoRA-DT address this issue by leveraging the transformer structure to store and transfer knowledge across tasks.
Continual Offline Reinforcement Learning via Diffusion-based Dual Generative Replay
A dual generative replay framework that leverages diffusion models to model states and behaviors with high fidelity, allowing the continual learning policy to inherit the distributional expressivity and mitigate catastrophic forgetting.
Offline Experience Replay for Continual Offline Reinforcement Learning
An agent can continuously learn new skills via a sequence of pre-collected offline datasets by using a novel offline experience replay (OER) method, which addresses the challenges of catastrophic forgetting and distribution shift.
Vector-Quantized Continual Diffuser (VQ-CD) for Continual Offline Reinforcement Learning with Diverse Task Spaces
VQ-CD leverages vector quantization to align diverse state and action spaces, enabling a diffusion-based model with selective weight activation to effectively learn and retain knowledge across a sequence of offline reinforcement learning tasks, even with varying state and action dimensions.
정렬된 공간에서 선택적 가중치 활성화를 통한 지속적인 오프라인 강화 학습 해결
VQ-CD는 양자화 공간 정렬(QSA)과 선택적 가중치 활성화(SWA)를 통해 다양한 작업 시퀀스에서 지속적인 오프라인 강화 학습(CORL) 문제를 해결하는 새로운 프레임워크입니다.