<?xml version="1.0" encoding="utf-8" standalone="yes"?><rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom" xmlns:content="http://purl.org/rss/1.0/modules/content/"><channel><title>强化学习 · 武宇新的学习与工程笔记</title><link>https://wyxbupt.me/tags/%E5%BC%BA%E5%8C%96%E5%AD%A6%E4%B9%A0/</link><description>记录 LLM、Agentic RL、通信与工程实践的个人技术博客。</description><generator>Hugo</generator><language>zh-CN</language><lastBuildDate>Mon, 03 Aug 2026 00:00:00 +0800</lastBuildDate><atom:link href="https://wyxbupt.me/tags/%E5%BC%BA%E5%8C%96%E5%AD%A6%E4%B9%A0/index.xml" rel="self" type="application/rss+xml"/><item><title>从 RLHF 到 Agentic RL：重新理解 RL 的探索本质</title><link>https://wyxbupt.me/posts/agentic-rl/rl-exploration-essence/</link><pubDate>Mon, 03 Aug 2026 00:00:00 +0800</pubDate><guid>https://wyxbupt.me/posts/agentic-rl/rl-exploration-essence/</guid><description>SFT 是拟合分布，RL 是奖励驱动下的行为探索；RLHF 为什么像 SFT，headroom 为什么决定 GRPO 能否生效。</description></item><item><title>从策略梯度到 PPO：目标函数、GAE 与裁剪机制</title><link>https://wyxbupt.me/posts/agentic-rl/understanding-ppo/</link><pubDate>Mon, 04 May 2026 00:00:00 +0800</pubDate><guid>https://wyxbupt.me/posts/agentic-rl/understanding-ppo/</guid><description>从策略梯度、优势函数和 GAE 出发，推导 PPO 的概率比与 clipped objective，并说明它为何仍是 on-policy 算法。</description></item></channel></rss>