<?xml version="1.0" encoding="utf-8" standalone="yes"?><rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom"><channel><title>Deep Learning Theory on Chen Kai Blog</title><link>https://www.chenk.top/zh/tags/deep-learning-theory/</link><description>Recent content in Deep Learning Theory on Chen Kai Blog</description><generator>Hugo</generator><language>zh-CN</language><lastBuildDate>Thu, 29 Sep 2022 09:00:00 +0000</lastBuildDate><atom:link href="https://www.chenk.top/zh/tags/deep-learning-theory/index.xml" rel="self" type="application/rss+xml"/><item><title>优化理论（十一）：非凸优化与鞍点逃逸</title><link>https://www.chenk.top/zh/optimization-theory/11-nonconvex-saddle-escape/</link><pubDate>Thu, 29 Sep 2022 09:00:00 +0000</pubDate><guid>https://www.chenk.top/zh/optimization-theory/11-nonconvex-saddle-escape/</guid><description>&lt;p>对于非凸函数 &lt;span class="math-inline">$f$&lt;/span>
，梯度下降法（GD）没有全局收敛保证。我们最多只能说 &lt;span class="math-inline">$\nabla f(x_t) \to 0$&lt;/span>
——即算法会收敛到一个&lt;strong>平稳点（stationary point）&lt;/strong>，而该点可能是局部极小值、鞍点，甚至是局部极大值。本文要探讨的问题是：&lt;strong>在什么条件下，我们能得出更强的结论？&lt;/strong>&lt;/p></description></item></channel></rss>