<?xml version="1.0" encoding="utf-8"?><feed xmlns="http://www.w3.org/2005/Atom" ><generator uri="https://jekyllrb.com/" version="3.10.0">Jekyll</generator><link href="https://yiding-w.github.io//feed.xml" rel="self" type="application/atom+xml" /><link href="https://yiding-w.github.io//" rel="alternate" type="text/html" /><updated>2026-04-09T07:48:16+00:00</updated><id>https://yiding-w.github.io//feed.xml</id><title type="html">Yiding Wang （王奕丁）</title><subtitle>personal description</subtitle><author><name>Yiding Wang</name><email>Blancokdb@gmail.com</email></author><entry><title type="html">PaST: Parametric Skill Transfer</title><link href="https://yiding-w.github.io//blog/past/" rel="alternate" type="text/html" title="PaST: Parametric Skill Transfer" /><published>2026-01-23T00:00:00+00:00</published><updated>2026-01-23T00:00:00+00:00</updated><id>https://yiding-w.github.io//blog/past</id><content type="html" xml:base="https://yiding-w.github.io//blog/past/"><![CDATA[<p>PaST (Parametric Skill Transfer) separates knowledge acquisition from reasoning skills by extracting a transferable skill vector from RL training and injecting it into SFT-adapted models. This makes continual adaptation more efficient and robust across domains.</p>

<p><img src="/images/PaST" alt="PaST overview diagram" style="max-width: 100%; border-radius: 12px; box-shadow: 0 8px 16px rgba(0, 0, 0, 0.08); margin: 1.25rem 0;" /></p>

<p>Links:</p>
<ul>
  <li><a href="https://arxiv.org/pdf/2601.11258">Paper</a></li>
  <li><a href="https://past-blog.notion.site/">Full blog on Notion</a></li>
</ul>]]></content><author><name>Yiding Wang</name><email>Blancokdb@gmail.com</email></author><category term="blog" /><summary type="html"><![CDATA[A short overview of PaST and how skill vectors enable fast knowledge adaptation.]]></summary></entry></feed>