<?xml version="1.0" encoding="utf-8" standalone="yes"?><rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom"><channel><title>Metrics | Jingshu Wang</title><link>https://jingshuw.org/tag/metrics/</link><atom:link href="https://jingshuw.org/tag/metrics/index.xml" rel="self" type="application/rss+xml"/><description>Metrics</description><generator>Wowchemy (https://wowchemy.com)</generator><language>en-us</language><copyright>'@copy;' Jingshu Wang 2026</copyright><lastBuildDate>Tue, 25 Nov 2025 00:00:00 +0000</lastBuildDate><image><url>https://jingshuw.org/images/icon_hua2ec155b4296a9c9791d015323e16eb5_11927_512x512_fill_lanczos_center_2.png</url><title>Metrics</title><link>https://jingshuw.org/tag/metrics/</link></image><item><title>New Commentary on the PDS Metric</title><link>https://jingshuw.org/post/pds-commentary-2025/</link><pubDate>Tue, 25 Nov 2025 00:00:00 +0000</pubDate><guid>https://jingshuw.org/post/pds-commentary-2025/</guid><description>&lt;p>We have posted a short commentary on Arxiv that examines the behavior of the &lt;strong>PDS&lt;/strong> evaluation metric used in the Virtual Cell Challenge:&lt;/p>
&lt;p>👉 &lt;a href="https://arxiv.org/pdf/2511.16954" target="_blank" rel="noopener">arXiv:2511.16954&lt;/a>&lt;/p>
&lt;p>The note discusses why early improvements in PDS can be unexpectedly difficult to achieve, and how scaling can lead to significant gains by making the usual ℓ₁/ℓ₂-based PDS behave more like a cosine-similarity–based score, which is often more discriminative.&lt;/p>
&lt;p>We hope this analysis may be helpful for researchers who are studying the metric or developing methods designed to perform well under PDS.&lt;/p></description></item></channel></rss>