arXivDaily arXiv每日学术速递 周一至周五更新
北美退伍军人节(Veterans Day),12 日无更新
arXiv 2106.04799cs.LG
收藏

数据高效强化学习的预训练表示

Pretraining Representations for Data-Efficient Reinforcement Learning

Université de Montréal(蒙特利尔大学);Mila(米拉人工智能研究所);HEC Montréal(蒙特利尔高等商学院);CIFAR(加拿大高等研究院);Microsoft Research(微软研究院)

Max Schwarzer, Nitarshan Rajkumar, Michael Noukhovitch, Ankesh Anand, Laurent Charlin, Devon Hjelm, Philip Bachman, Aaron Courville

摘要:

数据效率是深度强化学习的关键挑战。我们通过使用未标记的数据来预训练编码器来解决这个问题,然后对少量特定于任务的数据进行微调。为了鼓励学习表示,捕获底层MDP的各个方面,我们采用了潜在的动态建模和无监督的目标条件RL的组合。当限制在Atari游戏上的10万步交互(相当于两个小时的人类经验)时,我们的方法大大超过了之前将离线表示预训练与特定任务微调相结合的工作,并且与其他需要数量级更多数据的预训练方法相比毫不逊色。我们的方法在与更大的模型以及更多样化、与任务一致的观测数据相结合时表现出了特别的前景--在我们最好的环境中,在Atari上接近人类水平的性能和数据效率。我们在https://github.com/mila-iqia/SGI上提供与此工作相关的代码。

英文摘要:

Data efficiency is a key challenge for deep reinforcement learning. We address this problem by using unlabeled data to pretrain an encoder which is then finetuned on a small amount of task-specific data. To encourage learning representations which capture diverse aspects of the underlying MDP, we employ a combination of latent dynamics modelling and unsupervised goal-conditioned RL. When limited to 100k steps of interaction on Atari games (equivalent to two hours of human experience), our approach significantly surpasses prior work combining offline representation pretraining with task-specific finetuning, and compares favourably with other pretraining methods that require orders of magnitude more data. Our approach shows particular promise when combined with larger models as well as more diverse, task-aligned observational data -- approaching human-level performance and data-efficiency on Atari in our best setting. We provide code associated with this work at https://github.com/mila-iqia/SGI.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑