<?xml version="1.0" encoding="utf-8" standalone="yes"?><rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom"><channel><title>Post-Training | Yiqiao Jin CS PhD @ Georgia Tech</title><link>https://ahren09.github.io/tags/post-training/</link><atom:link href="https://ahren09.github.io/tags/post-training/index.xml" rel="self" type="application/rss+xml"/><description>Post-Training</description><generator>Hugo Blox Builder (https://hugoblox.com)</generator><language>en-us</language><lastBuildDate>Mon, 05 Oct 2026 00:00:00 +0000</lastBuildDate><image><url>https://ahren09.github.io/media/icon_hu_eee6347cbdb2cc3f.png</url><title>Post-Training</title><link>https://ahren09.github.io/tags/post-training/</link></image><item><title>Can Language Models Learn to Reject Their Own Bad Reasoning Steps?</title><link>https://ahren09.github.io/publication/preprint26_self_step_rejection/</link><pubDate>Mon, 05 Oct 2026 00:00:00 +0000</pubDate><guid>https://ahren09.github.io/publication/preprint26_self_step_rejection/</guid><description>&lt;h2 id="overview">Overview&lt;/h2>
&lt;p>Self-Step Rejection (SSR) trains a lightweight LoRA acceptance gate on a frozen generator. Training signals identify reasoning steps that cross a recoverability threshold, combining classification, same-prefix comparisons, and policy refinement. During inference, the model accepts a candidate step or resamples from the same prefix within a rejection budget, using its own acceptance gate to guide reasoning.&lt;/p>
&lt;h2 id="links">Links&lt;/h2>
&lt;ul>
&lt;li>
&lt;/li>
&lt;/ul></description></item></channel></rss>