What is the criteria for a ML paper to be published?[D]

The question that sparked this discussion is one of those things that sounds basic until you try to answer it honestly: what does it actually take to get a machine learning paper published? The thread appeared mid-April, almost certainly from someone standing at the entrance to academic ML research, trying to figure out whether the door opens from the inside or the outside. That confusion is not naivety — it's an accurate read of a process that has become genuinely inconsistent. A decade ago, the field was smaller, the conferences fewer, and novelty carried real weight on its own. Show something that hadn't been shown before, demonstrate it works, write it up. That was often enough. Then the field exploded. NeurIPS, ICML, and ICLR submission counts went from hundreds to tens of thousands. Review committees scaled in headcount but not necessarily in calibration. The result is a publication landscape where the criteria — officially — involve novelty, technical soundness, empirical rigor, and clarity of writing, but where — unofficially — the weight given to each shifts by venue, reviewer, subfield, and year. A reinforcement learning paper and a fairness paper are not judged by the same invisible rulebook, even when submitted to the same conference.

If confirmed as a genuine point of community confusion rather than an isolated question, this matters more than it might appear. The people asking "what gets published" are often early-career researchers, PhD students navigating their first submissions, or practitioners trying to bridge industry work into academic credibility. If they cannot get a straight answer — and the evidence suggests they genuinely cannot — then the pipeline that feeds new researchers into the field has a leak in it. The second-order effect is a quiet filter that favours people with access to well-connected advisors who know the unwritten rules, over people who are trying to reverse-engineer the system from first principles. That's not a diversity footnote. That's a structural problem that shapes which ideas get developed, which get abandoned, and ultimately which directions the field moves in. Venues that clarify their actual criteria, not just their official ones, have a real opportunity to recruit better submissions and better researchers.

Watch for whether this thread generates follow-up posts from senior ML researchers or programme chairs willing to speak plainly about what reviewers actually weight in practice — that kind of candid response would sharpen the picture considerably.

NewsHive monitors these sources continuously. All signal titles above link to the original reporting.

Intelligence by NewsHive. Need help navigating what this means for your business? Contact GeekyBee →