Some thoughts on the train back from EurIPS 2025

Published on 2025/12/08.

EurIPS 2025 was a blast! I was utterly impressed by the quality and professionalism of the conference, even more so for a first edition. This was made possible through the tireless work of the General Chairs, Søren Hauberg and Aasa Feragen, who deserve most of the credit for the success of the event. When the worse incident in the whole week is the light being flickery during a few minutes at a poster session, one can safely say the event ran smoothly. I had a great time, and this is the case of everyone I talked to. A special mention to the amazing buffets, and to the bespoke EurIPA beer tasting that concluded the main program!

As I am traveling back home, I felt inclined to share a few thoughts about this conference in particular, and decentralized conferences in general. My initial motivation for getting involved into decentralized conferences was climate. While this concern did not go away, I now feel that the discussion needs to be structured around a more nuanced and positive discourse. Going to local conferences need not be felt as missing out because prioritizing climate, but actually as having a great experience while also shaping more sustainable research practices.

Having touched upon the topic with a number of people throughout the week, I see an encouraging momentum in favor of decentralized conferences (arguably in a biased, yet already quite significant, population). This narrative is building around a convergence of arguments: in no specific order, family constraints, European sovereignty, visa issues, financial reasons, preference for smaller venues, geopolitical concerns, worries about Big Tech influence, European collaboration, and yes, climate. At the same time, the continuing growth of main conference venues makes me more and more dubious of counter-arguments (good luck for building meaningful connections with new people, that go beyond what you might get from a zoom call, among 29 thousand attendees!).

One thing that particularly struck me while at EurIPS was how the event was helping to shape a sense of community among European researchers. At EurIPS, I felt proud to be a European working on machine learning. I must say this is a welcome break of a frequent temptation towards self-defeatism which would lead us to believe that we are incapable of building things on our own. On the contrary, EurIPS was the opportunity to remind ourselves that we are actually doing great research and innovation on a wide variety of topics. To cite only one example, research on tabular foundational models is extremely promising, including in terms of economic impact, and Europe is at the forefront in this area. I was very excited to learn about the latest updates on that topic at the conference.

Let me stress how aware I am of the importance of being careful in this line of discourse, to avoid falling into Europe-first and nationalist rhetorics in the name of overcoming self-defeatism. Science is first and foremost a global endeavor, and it is paramount that we continue collaborating with colleagues from all over the world. This being said, it does not hurt to celebrate our achievements from time to time, and at the moment I believe we collectively have quite a margin before it becomes excessive.

I was also positively surprised of the relatively small share of LLM-related content in the scientific program. I believe this was a great opportunity to equilibrate the narrative (yes, there are still other things beyond finetuning LLMs) and to raise awareness (ah, so this is what people around me are working on!). A few topics which I was surprised to see appear so much during the conference are applications to science, optimization for ML, responsible ML, and small languages models.

EurIPS was also a great opportunity to meet new (and old) people from all over. Insightful discussions with people working in Switzerland, Italy, Germany, Austria, Denmark, Sweden, Finland, the United Kingdom, Israel, but also Canada and India (and obviously France!) come to mind as I write those lines. This felt nice, and this kind of discussion forums goes a long way in binding us together, especially at a time where our community is besieged by pressure against the reviewing system and scientific values. I think this also undermines one of the main arguments against decentralized conferences, namely that they would be too small and self-centered to foster scientific communication and cross-fertilization. EurIPS had 2200 attendees, which was the size of NeurIPS not so long ago, and gave ample opportunity for exposure to diverse scientific contents and interactions outside of one’s network of collaborators.

I was also struck by the vibrant community participation in the event, be it through the 18(!) workshops organized over a very short timeframe, or through great affinity events. A special shootout to WiML, which gave out travel grants to help some of our colleagues to travel to the conference, in coordination with the sister WiML event at NeurIPS.

Let me conclude by paraphrasing a few words from the keynote of Gaël Varoquaux that struck me: pick research topics where you own the narrative, while keeping in mind that this narrative shapes society, even more now than ever.