<?xml version="1.0" encoding="utf-8" standalone="yes"?><rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom"><channel><title>Convolution on 111qqz的小窝</title><link>https://111qqz.com/tags/convolution/</link><description>Recent content in Convolution on 111qqz的小窝</description><generator>Hugo -- gohugo.io</generator><language>zh-cn</language><managingEditor>hust.111qqz@gmail.com (111qqz)</managingEditor><webMaster>hust.111qqz@gmail.com (111qqz)</webMaster><copyright>© 2011-2026 111qqz</copyright><lastBuildDate>Sun, 20 Sep 2026 10:00:00 +0800</lastBuildDate><atom:link href="https://111qqz.com/tags/convolution/index.xml" rel="self" type="application/rss+xml"/><item><title>9 年后重温 CNN：剥掉算子细节后，真正留下了什么</title><link>https://111qqz.com/2026/09/convnet-architecture-thinking/</link><pubDate>Sun, 20 Sep 2026 10:00:00 +0800</pubDate><author>hust.111qqz@gmail.com (111qqz)</author><guid>https://111qqz.com/2026/09/convnet-architecture-thinking/</guid><description>&lt;p&gt;最近在重新过 MIT 6.S191 Lecture 3（卷积神经网络）。2017 年刚接触 CV 那会儿，CNN 算是吃饭的家伙，每天都在调。后来精力逐渐转到 ML Infra，成天跟 GPU 显存、通信拓扑和算子优化打交道，卷积网络的很多具体细节就慢慢生疏了——写个 &lt;code&gt;nn.Conv2d&lt;/code&gt; 时 weight 的四维形状到底怎么排、Kaiming 初始化的方差怎么推、感受野怎么算，冷不丁被问到，还得在脑子里卡壳一下。&lt;/p&gt;</description></item></channel></rss>