<?xml version="1.0" encoding="UTF-8"?><rss version="2.0"><channel><title>冷雪言</title><description>记录生活、工作与学习。</description><link>https://lenxueyan.pages.dev/</link><item><title>反向传播不是新算法，只是链式法则的记账方式</title><link>https://lenxueyan.pages.dev/blog/backprop-chain-rule/</link><guid isPermaLink="true">https://lenxueyan.pages.dev/blog/backprop-chain-rule/</guid><description>把计算图画出来，每个节点的局部导数乘上上游传来的梯度，就是全部。</description><pubDate>Tue, 22 Sep 2026 00:00:00 GMT</pubDate><category>学习</category><category>数学</category><category>机器学习</category></item><item><title>用线性代数看懂梯度下降</title><link>https://lenxueyan.pages.dev/blog/linear-algebra-gradient-descent/</link><guid isPermaLink="true">https://lenxueyan.pages.dev/blog/linear-algebra-gradient-descent/</guid><description>把梯度下降还原成&quot;在曲面上找最低点&quot;这件事，需要的数学其实只有投影和一点线性代数。</description><pubDate>Sun, 20 Sep 2026 00:00:00 GMT</pubDate><category>学习</category><category>数学</category><category>机器学习</category></item><item><title>王国之泪，以及&quot;自由度&quot;的代价</title><link>https://lenxueyan.pages.dev/blog/zelda-totk-freedom/</link><guid isPermaLink="true">https://lenxueyan.pages.dev/blog/zelda-totk-freedom/</guid><description>给你越多解决路径，能用的引导手段就越少。这一代的迷宫设计其实是拿这个代价换回来的。</description><pubDate>Fri, 18 Sep 2026 00:00:00 GMT</pubDate><category>生活</category><category>游戏</category><category>开放世界</category><category>设计分析</category></item><item><title>周报该写给谁看</title><link>https://lenxueyan.pages.dev/blog/who-is-your-weekly-report-for/</link><guid isPermaLink="true">https://lenxueyan.pages.dev/blog/who-is-your-weekly-report-for/</guid><description>大多数周报写砸，是因为作者在向&quot;在场的人&quot;汇报，而读者其实是三个月后什么都不记得的那个人。</description><pubDate>Fri, 11 Sep 2026 00:00:00 GMT</pubDate><category>工作</category><category>方法论</category><category>协作</category></item></channel></rss>