WM PAI workshop

Navigating the Confusion: Workshop Reviews Don't Guarantee Acceptance

A reviewer score of 8, alongside a 5 and a 4, and still facing rejection, that's a confusing outcome, especially when papers with similar scores from the same workshop also didn't make the cut.

4 min readMachine Learning

Workshop reviews have always been a murky signal, and the experience posted by the user who submitted to the WM PAI workshop at NeurIPS makes that painfully clear. Scoring an 8, a 5, and a 4 from reviewers, only to be rejected, is confusing enough. But discovering that the papers they reviewed, which averaged around a 7, were also rejected raises a more uncomfortable question: what exactly is a workshop acceptance threshold, and does anyone know where it is? Our take is blunt: if you are navigating this system for the first time, you are not misunderstanding the process; the process is simply not transparent. This is a recurring theme in academic publishing, as we explored in Navigating Your First TMLR Submission: What Happens After You Upload, where the gap between submission and official communication can feel just as disorienting.

The core issue here is that review scores are not a guarantee of acceptance, especially at workshops, which operate under different constraints than main conferences. Workshops often prioritize thematic fit, novelty, and space limitations over raw numerical scores. A paper with an 8 might be rejected if it doesn't align with the workshop's specific focus, while a paper with a 6 that sparks discussion might be welcomed. The fact that the user saw their own rejection and the rejection of papers they reviewed, both with respectable scores, suggests the organizers were making tight, selective choices. But the silence from the organizers, with decisions visible on OpenReview but no official email sent, is a failure of communication. It leaves authors in a limbo that erodes trust. Compare this to the experience of a student who secured a poster slot at the MusIML workshop, as detailed in Student needs travel funding for NeurIPS workshop poster, that author at least knew where they stood and could plan accordingly.

What this means for you, as a researcher submitting to workshops, is practical. First, never treat a high review score as a proxy for acceptance. A score of 8 is excellent, but it is not a contract. Second, expect delays and incomplete communication. Workshop organizers are often volunteer-driven, and the end of the NeurIPS cycle is chaotic. If you see a decision on OpenReview, consider that the official word, even if no email follows. But do not assume the system is broken just because it feels opaque. In many cases, the review scores are only one input; the program committee's discussion and the workshop's capacity constraints matter just as much. We touched on how novelty critiques can derail a submission in Navigating Novelty Critiques in Computer Vision Research, and that dynamic applies here too, fit and framing often outweigh raw marks.

The one concrete takeaway is this: if you are relying on workshops for feedback or a line on your CV, build a buffer. Submit to multiple venues, and do not treat a single workshop's decision as a definitive measure of your work's quality. The user's confusion about whether any papers were accepted at all is a symptom of a system that prioritizes anonymity over clarity. Until workshops standardize their notification processes, or at least send a mass email, you are better off checking OpenReview manually and planning as if the decision you see there is final. That is not ideal, but it is the reality. And if you are a workshop organizer reading this: send the email. It takes five minutes and saves hours of confusion.

From Machine Learning

I submitted a paper to the WM PAI workshop at NeurIPS and can now see the reviews and decision on OpenReview, but I didn't receive an official acceptance/rejection email.

My reviewer scores were 8, 5, and 4, all with confidence 4, and the paper was ultimately rejected.

Read the original at Machine Learning