BriefGPT.xyz
Nov, 2023
可验证的表示与高效规划用于部分可观察强化学习
Provable Representation with Efficient Planning for Partially Observable Reinforcement Learning
HTML
PDF
Hongming Zhang, Tongzheng Ren, Chenjun Xiao, Dale Schuurmans, Bo Dai
TL;DR
本研究旨在解决强化学习中部分可观察马尔可夫决策过程带来的性能下降问题,并通过对表示视图的利用提出了一种可行的强化学习算法,可在部分观测输入下实现比现有算法更高的性能,推动可靠强化学习在实际应用中的应用。
Abstract
In real-world
reinforcement learning
problems, the state information is often only partially observable, which breaks the basic assumption in Markov decision processes, and thus, leads to inferior performances.
partiall
→