
  <rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom">
    <channel>
      <title>Ray&#39;s Blog</title>
      <link>http://localhost:3000/blog</link>
      <description>Notes on machine learning, NLP, and related topics.</description>
      <language>zh-cn</language>
      <managingEditor>huangyrcn@gmail.com (Ray)</managingEditor>
      <webMaster>huangyrcn@gmail.com (Ray)</webMaster>
      <lastBuildDate>Mon, 15 Jun 2026 00:00:00 GMT</lastBuildDate>
      <atom:link href="http://localhost:3000/feed.xml" rel="self" type="application/rss+xml"/>
      
  <item>
    <guid>http://localhost:3000/blog/trace-paper-reading</guid>
    <title>TRACE: 通过 Adaptation-Aware Probing 发现 Task-Specific 参数，实现持续微调不遗忘</title>
    <link>http://localhost:3000/blog/trace-paper-reading</link>
    <description>KDD 2026 论文 TRACE 解读——将 continual fine-tuning 重新定义为 task-specific parameter discovery 问题，用一次短暂的 warm-start probe 找到每个任务的核心参数子集，然后只更新这些参数，从而在持续微调中有效缓解 catastrophic forgetting。</description>
    <pubDate>Mon, 15 Jun 2026 00:00:00 GMT</pubDate>
    <author>huangyrcn@gmail.com (Ray)</author>
    <category>KDD</category><category>continual-learning</category><category>LLM</category><category>fine-tuning</category><category>paper-reading</category>
  </item>

    </channel>
  </rss>
