<?xml version="1.0" encoding="utf-8" standalone="yes"?><rss version="2.0" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:atom="http://www.w3.org/2005/Atom" xmlns:sy="http://purl.org/rss/1.0/modules/syndication/" xmlns:media="http://search.yahoo.com/mrss/"><channel><title>Neural-Network on k4i's blog</title><link>https://k4i.top/zh/tags/neural-network/</link><description>Recent content in Neural-Network on k4i's blog</description><generator>Hugo -- gohugo.io</generator><language>zh</language><managingEditor>sky_io@outlook.com (K4i)</managingEditor><webMaster>sky_io@outlook.com (K4i)</webMaster><copyright>All content is subject to the license of &lt;a rel="license noopener" href="https://creativecommons.org/licenses/by-nc-sa/4.0/" target="_blank"&gt;CC BY-NC-SA 4.0&lt;/a&gt; .</copyright><lastBuildDate>Thu, 18 Jun 2026 10:00:00 +0800</lastBuildDate><atom:link href="https://k4i.top/zh/tags/neural-network/index.xml" rel="self" type="application/rss+xml"/><item><title>Activation Function：神经网络里那个很小但很关键的非线性</title><link>https://k4i.top/zh/posts/activation-functions-neural-networks/</link><pubDate>Thu, 18 Jun 2026 10:00:00 +0800</pubDate><author>sky_io@outlook.com (K4i)</author><atom:modified>Thu, 18 Jun 2026 10:00:00 +0800</atom:modified><guid>https://k4i.top/zh/posts/activation-functions-neural-networks/</guid><description>&lt;p&gt;Activation function 看起来只是神经网络层里一个很小的函数。真正重的计算通常是矩阵乘法：&lt;/p&gt;
&lt;p&gt;$$z = Wx + b$$&lt;/p&gt;
&lt;p&gt;然后逐元素套一个函数：&lt;/p&gt;
&lt;p&gt;$$a = \phi(z)$$&lt;/p&gt;
&lt;p&gt;很容易把 \(\phi\) 当成一个可以随便替换的名字：sigmoid、tanh、ReLU、GELU、SiLU、Mish，或者几百个变体。但 activation function 不是装饰。它决定了多层网络能不能表达非线性函数，梯度能不能传下去，hidden value 的分布是否稳定，以及为了很小的精度收益是否要付出明显的运行成本。&lt;/p&gt;</description><dc:creator>K4i</dc:creator><media:content url="https://k4i.top//images/posts/activation-functions-neural-networks/activation-function-icon.svg" medium="image"><media:title type="html">featured image</media:title></media:content><category>deep-learning</category><category>activation-function</category><category>neural-network</category><category>gradient-descent</category><category>AI</category></item></channel></rss>