Skip to content

Guide

Rejected from NeurIPS 2026: where the paper can go now

The calendar has always decided where a rejected NeurIPS paper goes. This cycle it decides against you: ICLR 2027 set its abstract deadline for 18 September 2026, six days before the NeurIPS decisions on 24 September, where in both previous cycles it fell the day after.

Published 14 August 2026

Rejected, so the paper is back in your hands. That is the one point in the cycle where reading the manuscript can still change the outcome. The dates you are working against are on the deadline board

6 days ICLR closes abstracts before NeurIPS decides 18 against 24 September 2026
82.5% of rejected papers are rejected again on a rerun below the 23% base rate
427 of 1,264 rejected papers later found published a floor: 680 traced at all
0.78 points lost when a prior rejection is disclosed randomised trial, 10-point scale

Read this before you plan the resubmission

The collision

ICLR shut the door six days early

Two cycles running, ICLR opened abstract registration the day after NeurIPS notified, which is why every guide names it next. This cycle the order is reversed.

NeurIPS decisions against the next ICLR abstract deadline
CycleNeurIPS decisionsICLR abstractsOrder
NeurIPS 2024 into ICLR 202526 September 202427 September 2024One day after: you knew, then registered.
NeurIPS 2025 into ICLR 202618 September 202519 September 2025One day after again, hence the folklore.
NeurIPS 2026 into ICLR 202724 September 202618 September 2026Six days before: register, then find out.

Read 14 August 2026: NeurIPS 2026, 2025, 2024; ICLR 2027, 2026, 2025; policy from ICLR author guide and NeurIPS 2026 handbook; withdrawals from the ICLR 2026 retrospective.

The paper deadline is the day after, and that is the trap

The ICLR 2027 paper deadline, 25 September, admits nothing without an abstract from 18 September. No place can be held either: placeholders are removed, the 2026 author guide says deleted, and no author joins after that date.

Withdrawing early is not deciding blind

Reviews landed 22 July and the rebuttal ran to 3 August, so you have had eight weeks with them, and 5,042 of ICLR 2026's 19,525 valid submissions were withdrawn.

The one thing this page will not rule on

Whether an abstract registered while NeurIPS still holds the paper is itself a parallel submission. Neither policy addresses it, so anyone who tells you is guessing, us included. Ask the chairs.

The calendar

Where the last cohort went, and what is open

Cortes and Lawrence traced the 1,264 papers NeurIPS 2014 turned down: the destinations were the venues whose deadlines fell shortly after, not the ones that fitted the work. Here is where each stands today.

Every outlet the 2014 rejects reached, and every route open now
Route2014 rejects found thereNext dateWhat to know
AAAI-27 main track72Closed 28 JulyTwo months early, and its Phase 1 rejections go out the day NeurIPS notifies.
AISTATS 202757Abstracts 29 SeptemberFive days after the decision, papers 6 October, and the site warns its dates may change.
ICML 202733No call publishedICML 2026 closed 28 January, which September does not reach anyway.
CVPR 202717Registration 10 NovemberPapers 16 November, and the call says that date will not be changed.
A later NeurIPS15No call publishedThe handbook says nothing about resubmitting a paper it rejected: silence, not permission.
JMLR14None, rollingTakes conference work only if you cite it, go deeper and extend the results substantively.
IJCAI14Not checkedA different claim from unpublished.
ICLR 202713Abstracts 18 SeptemberPapers 25 September, reachable only by withdrawing six days early.
UAI 202711No call publishedUAI 2026 closed 25 February.
NeurIPS 2026 workshops29 August, suggestedChosen four weeks before the decision, before you know you need it; notification 29 September. Workshops set their own dates. Non-archival, dual submission allowed.
ACL Rolling Review12 OctoberThree weeks after, and the cleanest confirmed window for an NLP paper; the cycle closes 23 December. The anonymity period is gone, and has been since February 2024.
TMLRNone, rollingTargets nine weeks; 2025 medians were 91 days short, 104 long, against 126 at NeurIPS. An accepted paper reaches ICLR 2027 by 18 December through the journal-to-conference track, 150 slots, no further review.
JAIRNone, rolling8 to 12 weeks, and disclosing a prior conference version is mandatory. First on its summary rejection list: papers rejected elsewhere without extensive revisions.
AAAI-27 Journal Track30 OctoberTakes work published in AIJ, JAIR, JMLR and others, so a journal paper still reaches a conference.
COLMNot checked

Tinted rows are reachable after the decision. Counts from Cortes and Lawrence, who traced 680 of the 1,264 rejected papers: 177 only on arXiv, 76 PDFs with no venue, 427 published elsewhere. Only outlets with ten or more appear, so a blank count means the outlet did not. Dates read 14 August 2026, the CVPR, AISTATS and ARR rows re-read 10 September, from NeurIPS workshops, journal-to-conference; ARR dates, anonymity; the TMLR FAQ and 2025 report, criteria on our TMLR guide; JAIR and JMLR; the AAAI-27 Journal and main tracks; aistats.org and its 2026 call; icml.cc; cvpr.thecvf.com; auai.org.

The authors named the mechanism themselves

Their note says there is likely a bias towards outlets with a deadline shortly after NeurIPS decisions are public: AAAI 2015 closed six days after they went out, AISTATS one month after. The top two destinations were the two venues open, and both have moved.

Where a date does not exist, this page does not print one

ICML 2027, UAI 2027 and NeurIPS 2027 have published no call for papers as of 10 September 2026, and no row above is filled in from last year. CVPR 2027 and AISTATS 2027 were in that state on 14 August; both now carry dates copied from the venue's own page, not from the aggregators already printing one.

The deadlines we have verified, with the date each was read

The number nobody quotes

The famous 50% is not about your paper

Every consolation post cites the finding that about half the accepted papers would have been rejected by a second committee. It is real, from the same table, and about accepted papers. Here is the row about a rejection.

The duplicate review experiment and two later findings
What was measuredValueWhose paper it describes
Accepted by one committee, rejected by the other49.5%Accepted papers. The number you were quoted, and not about you.
Rejected by one committee, rejected again by the other82.5%Rejected papers. This one is.
A rejected paper in an independent rerunabout 17.5%Yours, as the complement of the row above.
NeurIPS 2014 acceptance rate, for comparison23%Every submission, and the bar the row above sits below.
Penalty for disclosing an earlier rejection0.78 pointsYours again, wherever disclosure is required, including the journals above. 133 reviewers, 19 papers, interval 1.30 to 0.24, novices hardest.
The same experiment rerun in 2021, at five times the volumeno measurable changeNext year's version of you. Arbitrariness did not worsen with scale.

First four rows from Cortes and Lawrence, revisiting the NeurIPS 2014 experiment: 166 papers, two committees, 101 rejected by both. Reject precision was 82% and 83%, 82.5% on average; its Appendix C contradicts its own confusion matrix, and these are the body-text numbers. The penalty is from Stelmakh, Shah, Singh and Daume, Prior and Prejudice, an arXiv preprint of 30 November 2020 whose venue we did not verify: one study, not a law. The rerun is Beygelzimer, Dauphin, Liang and Vaughan; more on AI review against human review.

Resubmitting essentially unchanged is a worse-than-average bet

An independent rerun rejects a rejected paper about 82.5% of the time, so its chance is around 17.5%, against an acceptance rate of 23%. It enters the next round not as an average submission but as one a committee has already turned down.

Peer review is random enough that a good paper gets rejected, and nowhere near random enough that a rejection is noise you can wait out. So the target is not a venue but a number: 82% is what a real revision has to beat. If you want the objections read out before a committee reads them again, that is what a paid pre-submission review is for, and it acts on the part of this page about the paper rather than the part about dates.

Does it follow the paper

Three regimes for whether your reviews travel

Where you go next decides who reads the last round. One of them publishes it whether you like it or not.

Whether a rejection becomes public and reviews follow
VenueIs the rejection public?Do the reviews follow?
ICLRYes, permanently. Rejected and withdrawn submissions stay public under author names, deanonymised at notification or immediately on withdrawal.As a record. Unaccepted papers are non-archival and may go elsewhere modified or not, but OpenReview keeps the reviews, comments and every version.
NeurIPSOnly if you opt in, within two weeks of notification. Reviews are published for accepted papers; doing nothing keeps yours private.No. AAAI-24 ran a fast track that took them, for papers scoring at least 4.9; AAAI-27 has none.
ACL Rolling ReviewInside ARR, by design. The October cycle is a new submission, but an ARR resubmission needs a link to the previous one, a summary of changes and a point-by-point response to each weakness; new reviewers see the old ones only after filing their own.

Empty cell: a question we did not verify. AAAI-25 and AAAI-26 were not checked, so treat the fast track as absent rather than withdrawn. Read 14 August 2026 from the ICLR 2027 call and author guide, the NeurIPS 2026 handbook, the ARR call and author guidelines and the AAAI-24 fast track page.

The next committee will not know it was rejected

It reads the paper it is handed, and 82% is the number a real revision has to beat. Specialists read the revised draft against the guidelines of the venue you are aiming at now and hand back the objections, with locations, before a second committee files them.

What the reviews were probably telling you: why ML conference papers get rejected, measured on 697 decisions. The limits of the 50% figure, including ours: AI review against human review.

phdflow reads a finished draft against the guidelines of the venue you are aiming at and returns an accept or reject call with the objections behind it. Scored against 697 real decisions: AUC 0.91 and 96% accuracy on 297 ICLR 2025 papers, one false accept in 149 rejects. The full measurement. Read a finished review before you pay for one. Your first paper gets a free preview; the full report is 60 credits, €9. No account needed. Pricing.