<?xml version="1.0" encoding="utf-8" standalone="yes"?><rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom"><channel><title>Transfer Learning |</title><link>https://lelain.net/tags/transfer-learning/</link><atom:link href="https://lelain.net/tags/transfer-learning/index.xml" rel="self" type="application/rss+xml"/><description>Transfer Learning</description><generator>HugoBlox Kit (https://hugoblox.com)</generator><language>en-us</language><lastBuildDate>Wed, 10 Jun 2026 00:00:00 +0000</lastBuildDate><image><url>https://lelain.net/media/icon.svg</url><title>Transfer Learning</title><link>https://lelain.net/tags/transfer-learning/</link></image><item><title>When foundation models meet astronomical data</title><link>https://lelain.net/publications/fm-astro-journal/</link><pubDate>Wed, 10 Jun 2026 00:00:00 +0000</pubDate><guid>https://lelain.net/publications/fm-astro-journal/</guid><description>&lt;p&gt;Journal article currently under review. The preprint is available on SSRN.&lt;/p&gt;
&lt;p&gt;This paper is the extended version of the benchmark first presented at
and
, and of the DINOv2 study published at
. It compares head-only and full fine-tuning of visual foundation models against modern CNNs on Galaxy10 DECaLS, and discusses what this implies for the use of off-the-shelf foundation models in astronomy. Fine-tuned models are released on
.&lt;/p&gt;</description></item></channel></rss>