arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 0707.3087cs.ITcs.LGmath.IT
收藏

通用强化学习

Universal Reinforcement Learning

Massachusetts Institute of Technology(麻省理工学院);Columbia University(哥伦比亚大学);Stanford University(斯坦福大学)

Vivek F. Farias, Ciamac C. Moallemi, Tsachy Weissman, Benjamin Van Roy

摘要:

我们考虑一个代理与一个未建模的环境。在每一次,代理人进行观察,采取行动,并产生成本。它的行动可能会影响未来的观察和成本。目标是使长期平均成本最小化。我们提出了一种新的算法,被称为主动LZ算法,从Lempel-Ziv计划的通用数据压缩和预测的思想为基础的最优控制。我们建立,在主动LZ算法,如果存在一个整数$K$,使得未来是有条件的独立于过去给定的$K$连续的行动和观察的窗口,那么平均成本收敛到最优。石头剪刀布游戏的实验结果表明了该算法的有效性。

英文摘要:

We consider an agent interacting with an unmodeled environment. At each time, the agent makes an observation, takes an action, and incurs a cost. Its actions can influence future observations and costs. The goal is to minimize the long-term average cost. We propose a novel algorithm, known as the active LZ algorithm, for optimal control based on ideas from the Lempel-Ziv scheme for universal data compression and prediction. We establish that, under the active LZ algorithm, if there exists an integer $K$ such that the future is conditionally independent of the past given a window of $K$ consecutive actions and observations, then the average cost converges to the optimum. Experimental results involving the game of Rock-Paper-Scissors illustrate merits of the algorithm.

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。

↑