Guide
Where to submit your machine learning paper
The useful question is not which venue is best. It is which room will judge your paper on the thing it actually has. Get that wrong and you lose six months to a rejection that was decided before anyone read your results.
Published 11 August 2026
Have the abstract to hand? Paste it lower on this page and the matcher ranks the venues that publish work like it. Jump to the matcher
The short version
- The useful question is not which venue is best. It is which room will judge your paper on the thing it actually has.
- Name the asset, not the topic. A result beating a real baseline wants a top general conference; an idea ahead of its evidence wants a venue with a discussion phase; a dataset wants a benchmarks track; careful applied work wants the domain's own venue.
- Getting it wrong costs six months, not a rejection. A paper sent to a room that cannot score its best feature is judged on its second best, and that is decided before anyone reads the results.
- 22% of papers land where the outcome is a coin flip. 154 of the 697 decisions we measured sat in the band where the venue, not the work, settles it.
Start here
Name the strongest thing your paper has
Not the topic. The asset. Venues differ mostly in which kind of strength they know how to reward, and a paper sent to a room that does not score its best feature is being judged on its second best.
| If the strongest thing is | Send it where | Because |
|---|---|---|
| A result that beats a real baseline | A top general conference. | This is the case those venues are built to score, and the one where their reviewer instructions give you the most credit for evidence. |
| An idea that is ahead of its evidence | A venue with a real discussion phase, or a workshop first. | A contribution that needs arguing needs somewhere to be argued. Without that, thin evidence is simply thin evidence. |
| A dataset, benchmark or artifact | A dedicated datasets and benchmarks track. | Judged on construction, documentation and usefulness rather than on novelty, which is the axis a main track would mark it down on. |
| Careful work in an applied domain | The domain's own venue or a journal. | General ML venues score novelty heavily. A field that rewards being correct over being first will not read your paper as unambitious. |
| A negative or replication result | A venue that names it as in scope. | Some now solicit these explicitly. Sent anywhere else it reads as a paper whose method did not work. |
| A theoretical contribution | A theory-friendly venue or a journal. | Page limits and a reviewer pool tuned to empirical work are both against you at a general conference. |
Deliberately named by kind rather than by conference. Which specific venue fits depends on your subfield, and the tracks and their scope statements change every cycle. Read the current call for papers before you commit to any of these.
The decision
Four questions, in this order
Who do you want to read this in three years
Ask it first because it settles most cases on its own and it is the only one that is about the work rather than about the calendar. If the people who would build on this go to one venue, that is the answer, and everything below is about timing rather than about choice.
Is the evidence finished, or is the idea finished
These fail differently. An unfinished idea with strong numbers survives review; a finished idea with thin numbers gets a soundness objection that no amount of writing answers. If the evidence is not there, the honest options are to run the experiments or to take the paper somewhere its argument can be discussed, not to submit and hope.
What does the next deadline cost you
The real cost of waiting for a better-fitting venue is a few months. The real cost of a bad fit is the same few months plus a rejection, minus the reviews you would have got from people who understood the work. Waiting is usually cheaper than it feels at two in the morning, and abstract registration closes before the paper deadline at the big venues, which is the deadline people actually miss.
Have you read the call for papers, or the folklore
Scope statements, tracks, page limits and policies on dual submission and preprints change every cycle, and most of what you know about a venue is from the last time you submitted. The call is short. The folklore is confidently wrong, which is the same failure this site has had to correct on its own pages twice.
What you are walking into
The bars are not where the folklore puts them
Two things worth knowing before you pick. Acceptance rates at the big venues are lower than most people carry in their heads, and the difference between the venues is smaller than the difference between papers.
About three papers in four are turned down
NeurIPS 2025 accepted 5,290 of 21,575 valid submissions, 24.52%. ICLR 2026 accepted 5,355 of 13,763 decisions, 27.4%, down from about 31% in each of the two years before. Rejection is the ordinary outcome for competent work, not a signal about your paper.
NeurIPS sits higher, but not by much
Reading 494 real decisions from both venues with the same scoring, the median accepted NeurIPS paper scored 6.90 and the median accepted ICLR paper 6.70. Real, consistent, and small next to the spread inside either venue. Choosing between these two on difficulty alone is optimising the wrong variable.
One paper in five is decided by something you cannot see
Across the 697 decisions we measured, 154 landed in a band half a point wide where acceptance ran at 41.6%. In that band nothing recoverable from the manuscript separates the accepted from the rejected. If your paper is there, venue choice matters less than having a second venue ready.
The full comparison of the two bars · What happens inside that band
The options people skip
A top conference is not the only place this can go
Three routes get dismissed by reflex and are right more often than their reputation suggests.
A workshop, when the idea is early
You get a deadline, a room of people who care about exactly this, and feedback while the work can still change. At most venues a non-archival workshop does not stop you submitting the full paper later, but that rule is venue-specific and worth checking rather than assuming.
A journal, when the work is careful rather than fast
No fixed deadline, no single-shot decision, and a revision cycle instead of a rebuttal window measured in days. For applied and theoretical work that does not fit a novelty-first reviewer pool, this is often the better room rather than the consolation one.
The second-choice conference, decided now rather than later
Pick it before you submit, while you can still see the paper clearly. Deciding after a rejection means deciding while demoralised and against a deadline, which is how papers get rewritten that did not need rewriting.
Free tool, no account
Where should this paper go?
Paste the title and the abstract. It ranks the venues that publish this kind of work, with the reach and the article charge OpenAlex has on record.
Every venue we hold, ranked against your paper
The matcher reads what your paper is about and ranks the venues that publish that subject, rather than the ones you happen to remember. Free, no account, and nothing you paste is kept.
Conference figures from the NeurIPS 2025 programme chairs and the ICLR 2026 retrospective. The objections that end papers, sorted by whether they can still be fixed, are in why papers get rejected.
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.