arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2102.08291cs.LG
收藏

基于图结构代理模型的Meta强化学习

Model-based Meta Reinforcement Learning using Graph Structured Surrogate Models

University of Amsterdam(阿姆斯特丹大学)

Qi Wang, Herke van Hoof

摘要:

强化学习是解决序列决策问题的一种很有前途的范式,但数据效率低和跨任务泛化能力弱是现实应用中的瓶颈。基于模型的Meta强化学习通过学习动态和利用来自先前经验的知识来解决这些问题。在本文中,我们仔细研究了这个框架,并提出了一个新的基于爱普生采样的方法,该方法包括一个新的模型来识别任务动态以及摊销策略优化步骤。我们表明,我们的模型,称为图结构的代理模型(GSSM),优于国家的最先进的方法在预测环境动态。此外,我们的方法是能够获得高回报,同时允许在部署过程中快速执行,避免测试时间策略梯度优化。

英文摘要:

Reinforcement learning is a promising paradigm for solving sequential decision-making problems, but low data efficiency and weak generalization across tasks are bottlenecks in real-world applications. Model-based meta reinforcement learning addresses these issues by learning dynamics and leveraging knowledge from prior experience. In this paper, we take a closer look at this framework, and propose a new Thompson-sampling based approach that consists of a new model to identify task dynamics together with an amortized policy optimization step. We show that our model, called a graph structured surrogate model (GSSM), outperforms state-of-the-art methods in predicting environment dynamics. Additionally, our approach is able to obtain high returns, while allowing fast execution during deployment by avoiding test time policy gradient optimization.

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。

↑