- 1、本文档共5页,可阅读全部内容。
- 2、原创力文档(book118)网站文档一经付费(服务费),不意味着购买了该文档的版权,仅供个人/单位学习、研究之用,不得用于商业用途,未经授权,严禁复制、发行、汇编、翻译或者网络传播等,侵权必究。
- 3、本站所有内容均由合作方或网友上传,本站不对文档的完整性、权威性及其观点立场正确性做任何保证或承诺!文档内容仅供研究参考,付费前请自行鉴别。如您付费,意味着您自己接受本站规则且自行承担风险,本站不退款、不进行额外附加服务;查看《如何避免下载的几个坑》。如果您已付费下载过本站文档,您可以点击 这里二次下载。
- 4、如文档侵犯商业秘密、侵犯著作权、侵犯人身权等,请点击“版权申诉”(推荐),也可以打举报电话:400-050-0827(电话支持时间:9:00-18:30)。
- 5、该文档为VIP文档,如果想要下载,成为VIP会员后,下载免费。
- 6、成为VIP后,下载本文档将扣除1次下载权益。下载后,不支持退款、换文档。如有疑问请联系我们。
- 7、成为VIP后,您将拥有八大权益,权益包括:VIP文档下载权益、阅读免打扰、文档格式转换、高级专利检索、专属身份标志、高级客服、多端互通、版权登记。
- 8、VIP文档为合作方或网友上传,每下载1次, 网站将根据用户上传文档的质量评分、类型等,对文档贡献者给予高额补贴、流量扶持。如果你也想贡献VIP文档。上传文档
查看更多
arg minblog
About Subscribe
Nesterovs Punctuated Equilibrium
Ben Recht • Apr 3, 2017
Ed. Note: this post is cowritten with Roy Frostig.
Following the remarkable success of AlphaGo, there
has been a groundswell of interest in reinforcement
learning for games, robotics, parameter tuning, and
even computer networking. In a landmark new paper
by Salimans, Ho, Chen, and Sutskever from OpenAI,
the authors show that a particular class of genetic
algorithms (called Evolutionary Strategies) gives
excellent performance on a variety of reinforcement
learning benchmarks. As optimizers, the application of
genetic algorithms raises red flags and usually causes
us to close browser windows. But fear not! As we will
explain, the particular algorithm deployed happens to
be a core method in optimization, and the fact that this
method is successful sheds light on the peculiarities of
reinforcement learning more than it does about genetic
algorithms in general.
Evolution Strategies is Gradient Ascent
Let’s look at the Evolutionary Strategies (ES) algorithm
proposed in the paper. The goal is to maximize some
reward function R(x) where x is dimensional. Their
algorithm computes the reward function at small
perturbations away from its current state, and then
aggregates the returned function values into a new
state. To be precise, they sample a collection of n
random directions ϵ to be normally distributed with
i
mean zero and covariance equal to the identity. Then
the algorithm updates it state according to the rule.
n
α
= + R( + σ ) .
x x ∑ x ϵ ϵ
t+1 t t i i
您可能关注的文档
- MCB_6e_ch08 Post-transcriptional gene control英文清晰版.ppt
- MCB_6e_ch13 Moving proteins into membranes and organellesMCB_6e_ch13将蛋白质转移到膜和细胞器中.ppt
- MCB_6e_ch15 Signal transduction and short-term cellular responses英文清晰版.ppt
- MCB_6e_ch21 Cell birth, lineage, and deathMCB_6e_ch21细胞的出生、谱系和死亡.ppt
- Mobile Software Testing – Automated Test Case Design Strategies英文清晰版.pdf
- nature of precautionary英文清晰版.pdf
- Neural networks and backpropagation explained in a simple way用一种简单的方式解释了神经网络和反向传播.pdf
- Next Life, Let Me Be Your Mother英文清晰版.doc
- nielsen-technology-sentiments-report-april-2014尼尔森科技情绪报告2014年4月.pdf
- No Excuse没有任何借口英文清晰版.ppt
文档评论(0)