Brief IA

ML Conferences: Logistical and Scientific Challenges

🔬 Research·Tom Levy·

ML Conferences: Logistical and Scientific Challenges

ML Conferences: Logistical and Scientific Challenges
Key Takeaways
1Conferences in machine learning, such as ICML and NeurIPS, are essential for publishing research, but acceptance rates are low.
2Traveling to these events, often international, involves hidden costs and significant logistical challenges, such as time zone differences.
3Returning from a conference comes with a backlog in daily work, disrupting the research rhythm.
💡Why it mattersResearchers must balance the value of conferences with the time and energy they consume, impacting their overall productivity.
Le brief IA que lisent les pros

Le brief IA que les pros lisent chaque soir

Les 7 actus IA du jour, décryptées en 5 min. Gratuit.

Inclus dès l'inscription : notre sélection des meilleurs guides & comparatifs IA.

Choisis ton rythme

Gratuit · Pas de spam · Désabonnement en 1 clic

📄
Full Analysis

The Disadvantages of Traveling for Conferences

For those unfamiliar with the world of machine learning (ML) research conferences, here’s a brief overview of the situation. If you are conducting ML research, you generally want to publish your work as a paper at a conference. An interesting but unpublished work holds little value, so it often comes down to publish or perish. However, to publish a paper, you first need to find a suitable venue. In the field of ML, the main conferences are three leading events: ICML, NeurIPS, and ICLR.

The International Conference on Machine Learning (ICML) was first held in 1980. During its first decade, it was organized as a workshop; subsequently, it continued as a full-fledged conference. In 1987, the first Conference on Neural Information Processing Systems (NeurIPS) took place in Denver, USA. ICLR joined these ranks more recently. In 2013, researchers founded the International Conference on Learning Representations (ICLR).

To publish a paper at one of these conferences, you must go through the sometimes tedious process of peer review. First, you submit an anonymized version of your paper, meaning you remove names, affiliations, and other identifying information. This version is then reviewed anonymously by peers, ideally those more experienced in your research area. You receive their feedback, have the opportunity to clarify or improve the paper, and then wait another month or two. Depending on the quality of the initial submission and how you addressed the reviewers' concerns, your paper will either be accepted or rejected.

This is where it gets important: acceptance rates are low. Generally, only about 20% to 30% of submitted papers are accepted. It’s a competitive game. Thus, if you manage to publish a paper at one of these conferences, you can consider yourself lucky. You can then attend the conference itself, which brings me back to the subject of this article.

Getting to the Conference

Conferences are held annually in different locations around the world. Depending on where you live, you may have a considerable distance to travel. If you are based in China and the conference is in South Korea, you can get there relatively quickly. In contrast, if you are coming from Europe or the United States, you will likely spend most of the day traveling. The same applies in reverse when the conference is held in the United States.

My last visit to a major conference was ICLR 2024, which took place in Vienna. For me, it was quite convenient to get there. Last month — and this explains the delay in writing this monthly review — I attended ICML in South Korea. The conference itself was enjoyable, but the journey took a lot of time. And it’s not just the travel that matters. There’s also the organizational burden and, perhaps more importantly, the opportunity cost.

The Hidden Costs of Conference Travel

First, you need to book flights and hotels. Generally, you also have to adhere to certain per diem limits, which means you have a daily budget to stick to for accommodation. So, you need to look for hotels near the conference venue, check if they fit the budget, get approval, and then book them. The same goes for flights. In total, this can easily take half a day — sometimes a whole day.

Then, the actual travel begins… Getting to the airport takes time. Waiting at the airport takes even more time. The flight itself can last over ten hours. Finally, the trip from the destination airport to the hotel adds another hour or two. After about 24 hours, you arrive at your destination. Great.

But then, there’s the time zone difference. The conference is in a different time zone, so your body needs time to adjust — ideally while you are already supposed to be attending talks, meeting people, and presenting your work. On the way back, the same thing happens in reverse. You spend another day traveling home, followed by another period of time zone adjustment.

In total, you have now spent two or three days organizing and executing the trip. Added to that are the conference days, which can add another three to seven days. Altogether, the conference can take up about ten days. This would be acceptable if the rest of life simply paused during that time. Unfortunately, that’s not the case.

The Backlog Waiting for You at Home

Upon your return, your regular work is still there. Emails have piled up. Tasks have been waiting. Projects have continued without you — or, perhaps worse, have not progressed at all. That’s what happened to me. After my return, I first had to deal with my personal backlog. There were things to manage, tasks to redo, and several threads to pick back up. Additionally, I had to prepare a revision for a journal and a response for the conference.

The conference itself was over, but the disruption surrounding it was not. And that is, for me, the real downside of traveling for conferences. It’s not just the long flight or the time zone difference. It’s that interruption of your normal rhythm: the preparation beforehand, the travel itself, the time spent at the event, and the backlog waiting for you afterward.

All of this can easily kill a week or two of progress on a new research project. And an extra week or two on a paper can make a significant difference. It could be another proof, it could be coding a challenging experiment, drafting a complete manuscript. For all these things, you need extended and uninterrupted time, ideally over several consecutive days. This is what is referred to as deep work (a topic I am particularly interested in; see also my recent review of my last 8.5 years in machine learning).

This doesn’t mean that conferences are useless. No, conferences are valuable. You meet other researchers, discuss ideas, discover new work, and finally, put faces to names you may have only known through papers. Nevertheless, this value comes at a considerable cost in time and energy.

So, the lesson for this month was rather simple: when planning conference travel, don’t just count the days indicated in the conference schedule. Count the preparation, the travel, the recovery, and the work waiting for you upon your return. Otherwise, a five-day conference can easily consume two whole weeks without considering the opportunity costs.

Brief IA — L'actualité IA en français

L'essentiel de l'actualité de l'intelligence artificielle, décrypté et expliqué chaque jour.