<?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>线性探测 on 语音/音频论文速递</title>
    <link>https://nanless.github.io/audio-paper-digest-blog/tags/%E7%BA%BF%E6%80%A7%E6%8E%A2%E6%B5%8B/</link>
    <description>Recent content in 线性探测 on 语音/音频论文速递</description>
    <generator>Hugo</generator>
    <language>zh-cn</language>
    <lastBuildDate>Thu, 23 Apr 2026 00:00:00 +0000</lastBuildDate>
    <atom:link href="https://nanless.github.io/audio-paper-digest-blog/tags/%E7%BA%BF%E6%80%A7%E6%8E%A2%E6%B5%8B/index.xml" rel="self" type="application/rss+xml" />
    <item>
      <title>FLiP: Towards understanding and interpreting multimodal multilingual sentence embeddings</title>
      <link>https://nanless.github.io/audio-paper-digest-blog/posts/2026-04-23-flip-towards-understanding-and-interpreting/</link>
      <pubDate>Thu, 23 Apr 2026 00:00:00 +0000</pubDate>
      <guid>https://nanless.github.io/audio-paper-digest-blog/posts/2026-04-23-flip-towards-understanding-and-interpreting/</guid>
      <description>这篇论文旨在解决对多语言、多模态句子嵌入（如SONAR, LaBSE）的可解释性问题。核心方法是提出一种称为因子化线性投影（FLiP）的模型，通过将嵌入向量线性投影到词汇表空间来提取关键词，以此作为理解嵌入内容的代理任务。与之前非因子化的线性探测方法（如LiP）和SpLiCE相比，FLiP在关键词提</description>
    </item>
  </channel>
</rss>
