The numbers

ICLR 2023 ran May 1 to 5 at the Kigali Convention Centre, a venue that opened in 2016 and sits about five kilometres from the airport. It drew 3,758 participants from 73 countries. The programme came from 4,955 submissions with 1,575 accepted, an acceptance rate just under 32 percent, and the recorded content becomes free to the public on June 3 for anyone who could not attend either in person or through the hybrid track.

It was the first time one of the field's top-tier conferences met in Africa. The 2020 meeting was assigned to Addis Ababa and went online with everything else that year, so this was the first time the plan survived contact with the calendar. Before that the list of host cities reads as a tour of North America and Europe, with Vienna in 2021, and 2024 goes back to Vienna again.

What the location changed

The obvious effect is who could be in the room. For years the standard complaint from African researchers has been that the visa process for the United States, Canada, and the Schengen area treats a conference invitation as a reason for suspicion rather than a reason for entry, and that the rejection rate is high enough that many people do not apply. The conference's invitation letter system, which any registered attendee could generate from their account, and a host country that wanted the meeting, inverted that. Researchers who had never been able to present in person could do so, and researchers from the usual places found out what it is like to be the ones filling in visa forms.

We do not have a count of how many attendees came from African institutions, and we would want one before making claims about representation. What we can say from the sessions we attended is that the questions after talks came from a wider range of places than usual, and that the hallway conversations included people who had only ever met their collaborators on video calls. That is anecdote and we are labelling it as such.

What it cost

Holding a conference far from where most attendees live has a price, and the honest accounting includes it. Flights from North America and East Asia were long and expensive, and we do not have the in-person figures from earlier years to hand for a fair comparison, and we would want them before saying the location reduced turnout. Some labs sent fewer people. Some senior researchers attended remotely and their students went in their place, which is arguably a feature rather than a cost.

The hybrid format carried more weight than usual because of this. The organisers committed to remote participation options and to releasing videos a month after the meeting. Whether that is a good substitute for being present depends on what you attend conferences for. If it is the talks, the videos are fine. If it is the two-minute conversation at a poster that turns into a collaboration, no stream replaces it, and the people who could not travel this time were in the position African researchers have been in every other year.

Whether the map moves

The test of whether Kigali mattered is what the conference circuit does next, and the early answer is not encouraging. ICLR 2024 is already set for Vienna, which will be its third time hosting. Africa has hosted once. A single meeting is a gesture. A rotation is a policy.

There is a version of the argument that says the meeting was the point, that it demonstrated the venue works, the logistics hold, and the science is unaffected, so the next organising committee that considers Nairobi or Accra or Cape Town has evidence instead of a hypothetical. We hope that is how it gets used. The other version is that Kigali becomes the answer whenever someone asks about geographic diversity, cited for a decade as the time the field did the thing.

What we would want tracked

Two numbers would settle a lot of the argument and neither is hard to collect. The first is the share of authors of accepted papers who could attend their own presentation, by region, for Kigali and for Vienna next year. The second is the visa rejection rate reported to the organisers, which conferences have started asking about but rarely publish. If the Kigali numbers on both are markedly better for African and South Asian authors, the case for rotating the venue writes itself.

Until then the honest summary is that ICLR ran a full-sized conference on a continent that had never hosted one, the papers were the same papers, and a few thousand people found out that the barrier had always been the paperwork.

Sources

  1. ICLR 2023 conference site
  2. ICLR 2023 visa and travel information
  3. Wikipedia, International Conference on Learning Representations