人工智能培训

搜索

机器学习论文:知识转移是否总是有助于学习更好的政策?(Does Knowledge Transfer Always Help to Learn a Better

[复制链接]
~Limi多多~ 发表于 2019-12-9 13:38:48 | 显示全部楼层 |阅读模式
~Limi多多~ 2019-12-9 13:38:48 161 0 显示全部楼层
机器学习论文:知识转移是否总是有助于学习更好的政策?(Does Knowledge Transfer Always Help to Learn a Better Policy?)当学习强化学习问题的策略时,保存样本的关键方法之一是使用来自近似模型(例如其模拟器)的知识。但是,从近似模型进行的知识转移是否总是有助于学习更好的政策?尽管对转移强化学习进行了大量的实证研究,但对这个问题的答案仍然难以捉摸。在本文中,我们提供了一个强有力的否定结果,表明即使是对近似模型的全面了解也可能无法减少用于学习正确策略的样本数量。真正的模型。我们构建了一个强化学习模型的示例,并显示了有或没有知识转移的复杂性具有相同的顺序。从好的方面来看,在其他假设下,有效的知识转移仍然是可能的。特别地,我们证明了知道真实模型的(线性)基础会大大减少用于学习准确策略的样本数量。
One of the key approaches to save samples when learning a policy for areinforcement learning problem is to use knowledge from an approximate modelsuch as its simulator.However, does knowledge transfer from approximate modelsalways help to learn a better policy?Despite numerous empirical studies oftransfer reinforcement learning, an answer to this question is still elusive.In this paper, we provide a strong negative result, showing that even the fullknowledge of an approximate model may not help reduce the number of samples forlearning an accurate policy ofthe true model.We construct an example ofreinforcement learning models and show that the complexity with or withoutknowledge transfer has the same order.On the bright side, effective knowledge transferring is still possible underadditional assumptions.In particular, we demonstrate that knowing the (linear)bases of the true model significantly reduces the number of samples forlearning an accurate policy.机器学习论文:知识转移是否总是有助于学习更好的政策?(Does Knowledge Transfer Always Help to Learn a Better Policy?)
URL地址:https://arxiv.org/abs/1912.02986     ----pdf下载地址:https://arxiv.org/pdf/1912.02986    ----机器学习论文:知识转移是否总是有助于学习更好的政策?(Does Knowledge Transfer Always Help to Learn a Better Policy?)
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

~Limi多多~当前离线
新手上路

查看:161 | 回复:0

快速回复 返回顶部 返回列表