<?xml version="1.0" encoding="utf-8" standalone="yes"?><rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom"><channel><title>Foundation Models on Slow Data</title><link>https://shika-b.github.io/tags/foundation-models/</link><description>Recent content in Foundation Models on Slow Data</description><generator>Hugo</generator><language>en-us</language><lastBuildDate>Wed, 27 May 2026 13:44:57 +0100</lastBuildDate><atom:link href="https://shika-b.github.io/tags/foundation-models/index.xml" rel="self" type="application/rss+xml"/><item><title>One-shot tabular prediction</title><link>https://shika-b.github.io/blog/tabular_foundation_models/</link><pubDate>Wed, 27 May 2026 13:44:57 +0100</pubDate><guid>https://shika-b.github.io/blog/tabular_foundation_models/</guid><description>&lt;p>Most of the data that actually runs the world lives in spreadsheets. Hospital records, bank transactions, sensor logs,
messy CSV your colleague emailed you last Tuesday: it&amp;rsquo;s really rows and columns all the way down. While we have been
hearing quite a lot about language models writing poetry and code and image models painting cathedrals, the
humble tabular prediction task was left to older methods&lt;sup id="fnref:1">&lt;a href="#fn:1" class="footnote-ref" role="doc-noteref">1&lt;/a>&lt;/sup> for quite a long time. That changed a few years ago, with
the introduction of various tabular foundation models&lt;sup id="fnref:2">&lt;a href="#fn:2" class="footnote-ref" role="doc-noteref">2&lt;/a>&lt;/sup>.&lt;/p></description></item></channel></rss>