<?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>jepa on Sparse Notes</title>
    <link>https://sparsenotes.com/tags/jepa/</link>
    <description>Recent content in jepa on Sparse Notes</description>
    <image>
      <url>https://sparsenotes.com/images/og-default.png</url>
      <link>https://sparsenotes.com/images/og-default.png</link>
    </image>
    <generator>Hugo -- gohugo.io</generator>
    <lastBuildDate>Sat, 29 Aug 2026 00:00:00 +0000</lastBuildDate><atom:link href="https://sparsenotes.com/tags/jepa/index.xml" rel="self" type="application/rss+xml" />
    <item>
      <title>LeVJEPA: Efficient &amp; Scalable Video Pretraining without the Heuristics</title>
      <link>https://sparsenotes.com/posts/2026/08/papers/levjepa-video-pretraining/</link>
      <pubDate>Sat, 29 Aug 2026 00:00:00 +0000</pubDate>
      
      <guid>https://sparsenotes.com/posts/2026/08/papers/levjepa-video-pretraining/</guid>
      <description>Replacing EMA target encoders and predictors with a provable anti-collapse regularizer lets a single video encoder match V-JEPA 2 at 5.6–20.8x less compute — and discarding 95% of tokens improves accuracy rather than degrading it.</description>
    </item>
    
  </channel>
</rss>
