https://hackerpaper.kaipfister.de/p/2304.03674/Collected HN comments and submissions linking this paper.enMon, 02 Nov 2026 11:16:22 GMThttps://news.ycombinator.com/item?id=38772142urn:hackerpapers:mention:2304.03674:38772142Tue, 26 Dec 2023 14:41:28 GMT<p>An interesting approach I came across at NeurIPS a few weeks ago is called &quot;ML with Requirements&quot;[1]: https://arxiv.org/abs/2304.03674 My basic understanding is that it combines &quot;standard&quot; supervised learning techniques (neural nets + SGD) with a set of logical requirements (e.g. in the case of annotating autonomous driving data, things like &quot;a traffic light cannot be red and green at the same time&quot;). The logical requirements not only make the solution more practically useful, but can also help it learn the &quot;right&quot; solution with less labelled data. [1] I don&#x27;t know if they had a NeurIPS paper…</p><p><a href="https://hackerpaper.kaipfister.de/p/2304.03674/">Paper and discussion history</a>. A mention is not an endorsement.</p>cs.LG