Hiring NeurIPS, ICML and ICLR authors

What acceptance certifies, and what it leaves open

A paper at NeurIPS or ICML went through several reviewers who read it against a stack of competing submissions and wrote down why it should appear. Most submissions are turned away. That process is the closest thing this field has to an external check on whether a piece of work holds up, and it happens before anyone is hiring.

The limit is that acceptance belongs to the paper rather than to any one person on it. A method paper out of a large lab can carry eight names, and the eighth may have contributed a week of compute and a figure. Reading the venue without reading the author list is how a shortlist fills up with people who were in the room.

Venue by venue

Each venue selects for something different, and the counts below describe how large each community is inside this index rather than how selective it is. The index tracks 21 of them. AAAI is the clearest example of why size and standing come apart: it publishes several times more papers than NeurIPS, which carries its author count to 58 172 while its reputation among research hires sits behind.

For method and theory work, NeurIPS and ICML carry the most weight. ICLR attracts representation learning and anything about training at scale. Vision sits at CVPR, ICCV and ECCV. Language work runs through ACL, which accounts for 16 514 authors here, along with EMNLP and NAACL. Robot learning sits at CoRL. KDD is worth reading when the role involves applied systems running on real data, because its papers tend to describe something that shipped.

The number that separates candidates

Once you have a pool of authors, first-authored papers at reviewed venues does more sorting than any other single field. It answers a question hiring managers actually care about, which is whether the person has carried a piece of research from an idea to something strangers accepted.

Total publication count pulls in the opposite direction. Someone in a large group appears on everything the group produces, so the count climbs with the size of the lab. Citation count has a different problem, described in our guide on finding and hiring AI researchers: citations take two to four years to arrive, so the researchers about to become available are exactly the ones who look weakest by that measure.

When these authors answer

Doctoral students are the part of this population you can actually hire, and they are reachable during a narrow period. Write six to nine months before the defence. Earlier than that, the honest answer is that it is too soon to say. Later, the postdoc letter or the industry offer is signed and a message changes nothing.

Conference calendars give you a second, smaller opening. The weeks after a camera-ready deadline are the weeks when someone has a finished result they are willing to talk about and a little room in the schedule. A message that names the paper and asks a real question about it lands very differently in that window.

Across this index, 9 702 researchers have a doctorate end date stated by a public source, 863 of them in 2026 and 291 in 2027. Between 4 and 5 years separate a first indexed publication from the end of a doctorate, which is the arithmetic behind knowing when to write to someone whose first paper appeared three years ago.

Writing to an author

Researchers get contacted about their work constantly and have become good at spotting a template. Name the paper, say which part of it matters for the problem you are solving, and ask something only that person can answer. Two specific sentences will do more than a page of company description.

Where the address comes from matters as well. Most authors print a contact address on the first page of the paper or on a lab page, and reading it from a document they published is both reliable and lawful. An address generated from a name and an institution pattern is a guess, and a bounced first message costs you the introduction.

Common questions

How many researchers have published at NeurIPS?

This index holds 18 635 researchers with at least one NeurIPS paper, across 6 159 NeurIPS papers. ICML accounts for 21 461 authors, ICLR for 17 243 and CVPR for 15 868. A researcher appears under every venue they published at, so those four numbers overlap and should be read one at a time.

Does an accepted paper at NeurIPS or ICML prove someone is a strong researcher?

It proves the work cleared a review process that turns most submissions away, which puts it well ahead of a job title as evidence. The gap is attribution. Eight names on a method paper do not describe eight equal contributions, and the sixth of them may have run a week of compute. Check where the person sits in the author list before you weigh the venue.

Why does author order matter more than the venue itself?

In machine learning the first author led the work and wrote most of it. Counting first-authored papers at reviewed venues tells you how many accepted pieces of research a person has driven. Counting total publications rewards sitting in a large group where every member appears on everything.

Which venue should I look at for the role I am filling?

NeurIPS and ICML for method and theory, ICLR for representation learning and training at scale, CVPR, ICCV and ECCV for vision, ACL, EMNLP and NAACL for language, CoRL for robotics, KDD for applied systems on real data. AAAI publishes far more papers than the others, so its author count is larger without the bar being higher.

When can I reach someone who just published at one of these venues?

The window that matters is the six to nine months before a doctorate ends. A first author on a recent NeurIPS or ICML paper who is still a doctoral student is fielding postdoc and industry offers in that period, and academic offers go out between January and April. After the camera-ready deadline the person has a finished result to talk about and a few free weeks, which makes it a good moment to write.

Can I get the list of authors at a given venue?

Yes. The index is searchable by subject and by venue, with first-authored counts, affiliation country, career stage and, where a public source states it, the year the doctorate ends. Access is by request.

Do you recruit these people, or only sell the data?

Both. The index is available on its own, and our recruitment agency runs the search from the same data with a fee paid when someone joins.