dyweb / papers-notebook

:page_facing_up: :cn: :page_with_curl: 论文阅读笔记(分布式系统、虚拟化、机器学习)Papers Notebook (Distributed System, Virtualization, Machine Learning)
https://github.com/dyweb/papers-notebook/issues?utf8=%E2%9C%93&q=is%3Aissue+is%3Aopen+-label%3ATODO-%E6%9C%AA%E8%AF%BB
Apache License 2.0
2.14k stars 250 forks source link

Automated Hyperparameter Tuning for Effective Machine Learning #76

Open gaocegege opened 6 years ago

gaocegege commented 6 years ago

https://pdfs.semanticscholar.org/ce60/a2a90be9ba9088f91b1bd51a2c1cabade1ae.pdf

个人觉得这篇文章不如 Google Vizier 抽象的好,做的事情几乎是一模一样的

at15 commented 6 years ago

难得策策又开始看 paper

gaocegege commented 6 years ago

瞎说,只是放上来而已

gaocegege commented 6 years ago

这篇文章的相关工作介绍做的很好,介绍了常见的几种超参数搜索的方法:

此外,还总结了超参数训练的一些挑战所在:

当然,还有更大的问题是训练成本。

这篇论文的核心就是一张图,了解一下

screenshot from 2018-05-02 12-00-03

xplorld commented 6 years ago

神奇,参数对效果不可微的话看起来只能画格子暴搜?

gaocegege commented 6 years ago

Optimization 类的算法用的好像是比较多的

xplorld commented 6 years ago

sgd 么

gaocegege commented 6 years ago

贝叶斯优化

xplorld commented 6 years ago

(思考)