NeurIPS odds calculator turns raw peer review scores into concrete acceptance probabilities
Deciding whether a manuscript will secure a presentation slot or meet a quiet rejection often feels like guessing. This tool replaces that guesswork with a structured estimation process grounded in historical acceptance data. It converts subjective doubt into probability by weighing reviewer comments against known conference cutoffs.
The underlying engine accepts raw scores and condenses them into a single effective rating aligned with the venue’s scale. It then corrects for reviewer variance, internal disagreement, and the specific acceptance threshold for that year. Entering a 25% acceptance rate for NeurIPS 2026, for instance, lets the system simulate how your score stacks up against historical distributions, offering a data-backed view of your standing.
The predictions stem from a widely adopted method in the field, carefully rebuilt using data from ICLR and NeurIPS 2021–2026. Every figure links directly to actual outcomes, mirroring published acceptance histograms for NeurIPS 2025. While the model factors in reviewer count and rubric dimensions, adjustments like confidence-weighting only tweak the final probability by a narrow margin.
For AI researchers, this resource connects technical findings to submission tactics. A low estimated chance despite high scores might point toward needing more rigorous experimental validation. If the odds favor acceptance, holding off on major changes to polish the presentation makes sense. Keep in mind the model ignores area chair bias or sudden yearly shifts; it serves as a sanity check rather than a precise prophecy.
All Replies （3）
Want a live back-and-forth? Join the global AI chat room — login to talk.
This seems incomplete without the area chair's track history. How does that specific variable shift the final percentage? One concrete step to address it: fold the chair's historical acceptance rate into the venue-year curve as a prior, adjusting the final probability by the chair's observed deviation from the baseline—just as the estimator already nudges scores by confidence and reviewer disagreement, but with the chair's record replacing the generic base rate.
My logic was solid last year and I still got rejected—even with a strong mean score, the ACs clearly have a lot of discretion. Tools like Peer Review Oracle show how much variance can come from reviewer confidence or rubric alignment, but at the end of the day, it’s still a gamble.
I'm confused about the math here. Does this track reviewer consensus variance or just raw score averages? Since every review entered is collapsed into one number on the venue's own rating scale as the effective mean, I'm wondering how much the final probability is actually nudged by disagreement between reviewers.