<?xml version="1.0" encoding="utf-8" standalone="yes"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom">
  <channel>
    <title>MoE on BOLIN的个人博客</title>
    <link>https://example.org/tags/moe/</link>
    <description>Recent content in MoE on BOLIN的个人博客</description>
    <generator>Hugo -- gohugo.io</generator>
    <language>zh-cn</language>
    <copyright>© 2026 BOLIN</copyright>
    <lastBuildDate>Wed, 29 Jul 2026 00:00:00 +0000</lastBuildDate><atom:link href="https://example.org/tags/moe/index.xml" rel="self" type="application/rss+xml" />
    
    <item>
      <title>FoundIR-v2：统一图像恢复也是一个资源调度问题</title>
      <link>https://example.org/papers/foundir-v2/</link>
      <pubDate>Wed, 29 Jul 2026 00:00:00 +0000</pubDate>
      
      <guid>https://example.org/papers/foundir-v2/</guid>
      <description>FoundIR-v2 的两个核心创新并不复杂：在输入端动态调整各任务的数据比例，在模型端利用 MoE 分配任务自适应扩散先验。真正值得关注的是，它把统一图像恢复重新理解成了一个资源调度问题。</description>
      
    </item>
    
    <item>
      <title>UniLDiff：把已有模块放到正确的位置</title>
      <link>https://example.org/papers/unildiff/</link>
      <pubDate>Wed, 29 Jul 2026 00:00:00 +0000</pubDate>
      
      <guid>https://example.org/papers/unildiff/</guid>
      <description>DAFF 负责在扩散轨迹中对齐 LQ 条件，DAEM 负责在 VAE 解码阶段补偿细节。相比 FoundIR-v2，UniLDiff 的研究问题更具体，也更偏网络结构工程。</description>
      
    </item>
    
  </channel>
</rss>
