<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom"><channel><title>On-the-Fly Adaptation of Source Code Models using Meta-Learning · HackerPapers discussions</title><link>https://hackerpaper.kaipfister.de/p/2003.11768/</link><description>Collected HN comments and submissions linking this paper.</description><language>en</language><lastBuildDate>Mon, 02 Nov 2026 11:16:22 GMT</lastBuildDate><atom:link href="https://hackerpaper.kaipfister.de/p/2003.11768/feed.xml" rel="self" type="application/rss+xml" /><item><title>On-the-Fly Adaptation of Source Code Models using Meta-Learning — mentioned by brzozowski</title><link>https://news.ycombinator.com/item?id=23883583</link><guid isPermaLink="false">urn:hackerpapers:mention:2003.11768:23883583</guid><pubDate>Sat, 18 Jul 2020 18:28:38 GMT</pubDate><description>&lt;p&gt;I would say that’s becoming increasingly plausible, but it’s not exactly what the authors show in this paper. In order to translate from Datalog queries, you would need to show how to encode arbitrary propositional formulae as a graph reachability problem. This appears to be possible following Reps et al. [1], but is still far away from becoming a push-button solution. Here, they are proposing a new architecture which is capable of inferring abstract relations between program states (e.g. variables), trained on a synthetic dataset of pairwise relations, and empirically showing its…&lt;/p&gt;&lt;p&gt;&lt;a href="https://hackerpaper.kaipfister.de/p/2003.11768/"&gt;Paper and discussion history&lt;/a&gt;. A mention is not an endorsement.&lt;/p&gt;</description><category>cs.LG</category></item></channel></rss>
