人工智能培训

搜索

人工智能论文:通过潜在状态解码具有丰富观察能力的高效RL(Provably efficient RL with Rich Observations via La

[复制链接]
aiwen 发表于 2019-1-28 11:22:24 | 显示全部楼层 |阅读模式
aiwen 2019-1-28 11:22:24 289 0 显示全部楼层
人工智能论文:通过潜在状态解码具有丰富观察能力的高效RL(Provably efficient RL with Rich Observations via Latent State Decoding)我们研究了具有从少数潜在状态产生的丰富观察的情节MDP中的探索问题。在某些可识别性假设下,我们演示了如何通过一系列回归和聚类步骤来诱导地估计从观察到状态的映射 - 其中先前解码的潜在状态为后来的回归问题提供标签 - 并用它来构建良好的勘探政策。我们对学习状态解码函数和探索策略的质量提供有限样本保证,并通过对一类硬探索问题的经验评估来补充我们的理论。我们的方法通过天真的探索成倍地改善了超过$ Q $ -learning,即使$ Q $ -learning已经进入潜伏状态。
We study the exploration problem in episodic MDPs with rich observationsgenerated from a small number of latent states.Under certain identifiabilityassumptions, we demonstrate how to estimate a mapping from the observations tolatent states inductively through a sequence of regression and clusteringsteps---where previously decoded latent states provide labels for laterregression problems---and use it to construct good exploration policies.Weprovide finite-sample guarantees on the quality of the learned state decodingfunction and exploration policies, and complement our theory with an empiricalevaluation on a class of hard exploration problems.Our method exponentiallyimproves over $Q$-learning with naïve exploration, even when $Q$-learning hascheating access to latent states.人工智能论文:通过潜在状态解码具有丰富观察能力的高效RL(Provably efficient RL with Rich Observations via Latent State Decoding) ZdjzDl1E38Z87Jj1.jpg
URL地址:https://arxiv.org/abs/1901.09018     ----pdf下载地址:https://arxiv.org/pdf/1901.09018    ----人工智能论文:通过潜在状态解码具有丰富观察能力的高效RL(Provably efficient RL with Rich Observations via Latent State Decoding)
回复

使用道具 举报

您需要登录后才可以回帖 登录 | 立即注册

本版积分规则 返回列表 发新帖

aiwen当前离线
新手上路

查看:289 | 回复:0

快速回复 返回顶部 返回列表