<?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>Contrastive Learning on Yukari's Blog</title><link>https://yukar.icu/tags/contrastive-learning/</link><description>Recent content in Contrastive Learning on Yukari's Blog</description><generator>Hugo</generator><language>zh-cn</language><lastBuildDate>Sun, 05 Jul 2026 00:00:00 +0000</lastBuildDate><atom:link href="https://yukar.icu/tags/contrastive-learning/index.xml" rel="self" type="application/rss+xml"/><item><title>Siglip2 损失函数详解 — Sigmoid Loss 如何评估图文对齐</title><link>https://yukar.icu/posts/vlm/siglip2/siglip2-loss-function-analysis/</link><pubDate>Sun, 05 Jul 2026 00:00:00 +0000</pubDate><guid>https://yukar.icu/posts/vlm/siglip2/siglip2-loss-function-analysis/</guid><description>从 modeling_siglip2.py 源码出发，完整推导 Siglip2 的 Sigmoid Contrastive Loss。 与 CLIP 的 Softmax CE Loss 逐行对比，解释可学习温度与偏置的设计动机。</description></item></channel></rss>