NeurIPS 2026 Splits Across Sydney, Atlanta, and Paris: What a Three-City Conference Means for Research Access
NeurIPS 2026 runs a main venue in Sydney with official satellites in Atlanta and Paris. What the multi-site format changes for authors, attendees, and access.
By Institute for Joint Cognition & AI · 6 min · 20 October 2026
NeurIPS 2026 is not one conference in one city. The organizers list the main event in Sydney, Australia, from December 6 to 12, with two official satellites in Atlanta, Georgia and Paris, France, both running into December 13 (NeurIPS 2026 conference page). For a field whose flagship venue has spent years outgrowing convention centres and running into visa walls, the format is the most significant structural change to the meeting in a long time. It is also an experiment whose costs and benefits will land unevenly depending on where a researcher lives, what they are presenting, and how early they registered.
How the three sites are arranged
The arrangement is a main venue plus satellites, not three equal co-hosts. Sydney carries the full program across seven days. Atlanta and Paris open slightly later and close a day after Sydney. The official site lists both satellites from December 9; the organizers’ own April 2026 newsletter gave Atlanta a December 8 start. Anyone booking travel should check the conference page directly rather than relying on secondary listings.
According to that newsletter, both satellite events will carry keynotes, oral and poster presentations of NeurIPS 2026 papers, and workshops, with tutorials, affinity events, and other elements still being planned for the satellite sites at the time of writing. The satellites are therefore presentation venues in their own right, not viewing rooms for a Sydney livestream.
Workshops by city
The August 2026 newsletter reports 477 workshop proposals (454 valid) and 102 accepted workshops, divided as follows:
| Site | Accepted workshops | Acceptance rate | Workshop days |
|---|---|---|---|
| Sydney | 48 | 21.5% | Dec 11–12 |
| Paris | 28 | 25.4% | Dec 12–13 |
| Atlanta | 26 | 23.6% | Dec 12–13 |
Sydney hosts slightly under half of the workshops; the two satellites together host slightly more than half. Organizers stated that, apart from a handful of workshops placed at their second-choice site, primary venue preferences were met. For researchers who follow a subfield through its workshop rather than the main track, this split matters more than the main-venue designation: the workshop you care about may not be in Sydney at all.
Where the format came from
The 2026 design builds on a 2025 pilot. NeurIPS 2025 ran its main conference in San Diego alongside a smaller official satellite in Mexico City, described by the organizers as a gathering of roughly 500 people, and separately endorsed EurIPS in Copenhagen as an additional venue where authors could choose to present papers that had been accepted at NeurIPS (NeurIPS 2025 Mexico City registration details). The stated motivations for the pilot were rising attendance and the visa difficulties some attendees had faced when only one location was offered.
Neither pressure has gone away. Growth strains any single venue, and the burden of obtaining a visa for any single host country falls hardest on researchers from regions with the most restrictive travel regimes. Choosing Sydney as the main venue shifts the default travel burden toward Asia-Pacific; adding Atlanta and Paris gives North American and European researchers a shorter trip.
How registration actually works
The access question is decided by the registration rules, which the organizers set out in a multi-site registration explainer published in August.
Authors
For the core tracks (Main, Evaluations & Datasets, Position Papers, Journal-to-Conference, and the new Reproducibility track), each paper that is accepted gets one guaranteed registration, reserved for a designated presenting author. Authors stated a preferred presenting site before decisions were made, and the organizers say that preference had no bearing on acceptance. Where a site’s capacity is exceeded, remaining slots are allocated through a randomized process that weighs necessity and preference. Once allocations are final, the presenter and the location cannot be changed.
Two practical consequences follow. Co-authors who are not the designated presenter register as members of the general public, with no guaranteed seat. And a presenter who registered early for the wrong site has to cancel and re-register at the assigned one; the organizers set a full-refund window for that until November 4.
Workshops, competitions, Creative AI, and Education tracks receive one reserved registration per accepted contribution, tied to the site where it will be presented, provided organizers import contributions before the event platform’s deadlines.
Everyone else
General public registration is first-come, first-served rather than a lottery. Each site keeps its own waitlist, a person may join only one, and any additional capacity is filled exclusively from those waitlists in a randomized order weighted toward original position. Sydney and Atlanta registration opened on September 3; the August newsletter said Paris registration was still undergoing separate testing, with an opening date to be announced.
What changes for research access
The multi-site format solves some access problems and introduces others.
Gains
- Shorter, cheaper travel for many attendees. A researcher in Western Europe can reach Paris by train; one in the eastern United States can reach Atlanta without a long-haul flight. That lowers the cost of attending for students and early-career researchers without large travel budgets.
- More than one visa path. A presenter who cannot obtain an Australian visa in time has two alternative jurisdictions, which was the stated purpose of the 2025 pilot.
- Distributed capacity. Spreading attendance across three venues eases the capacity limit of any single building.
Costs
- A fragmented hallway. Much of a conference’s value lies in unplanned conversations. A researcher in Paris will not meet the Sydney poster session, and the same paper’s authors may be split across sites.
- Workshop geography decides attendance. With 54 of 102 workshops outside Sydney, following two communities may now require choosing between them rather than walking between rooms.
- Co-author exclusion risk. Only the designated presenter has a guaranteed place. Groups with several student co-authors face first-come, first-served registration for the rest, and only one waitlist per person.
How to read the experiment
The 2025 pilot tested whether a satellite could work; 2026 makes satellites a formal part of the program, with presentations, workshops, and allocated registrations at each site. Its own reproducibility track, new as an official track this year, will present in the same multi-site structure.
The useful questions to ask after December are concrete: what share of attendance went to each site, whether satellite presenters drew comparable audiences, how many authors received their first-choice venue, and whether participation from regions that faced visa barriers rose. The organizers have published the inputs; the outcomes will show whether a three-city conference widens access or simply redistributes it. Researchers planning December travel should confirm their assigned site, register co-authors early, and check workshop locations before booking anything.