Why I am not attending NeurIPS 2023 in person

Published on 2023/10/03.

tl;dr: As a PhD student based in France, I will not be attending NeurIPS 2023 in New Orleans because the negative impact of air travel on the environment makes flying to conferences unsustainable. I am open to presenting one or both of my accepted NeurIPS papers in any gathering taking place in Europe around conference week.

My name is Pierre Marion, I am currently finishing my PhD at Sorbonne University in Paris, and I have two first-author accepted papers at NeurIPS 2023. I have never attended a major ML conference in person, since my only previously accepted paper in a major ML conference (NeurIPS 2021) was during the COVID-19 pandemic lockdowns. I love travelling and going to conferences, and I have funding to go to NeurIPS. Additionally, this conference in particular will feature workshops that are really interesting to me and I feel that I would benefit much—both academically and socially—from attending. Despite all of this, I have decided not to attend NeurIPS 2023 in person. This was a difficult decision to make, and my goal in this document is to explain the underlying reasoning in the event that this is of immediate interest to other researchers with the same hesitations, who I hope will join me in my decision. This note will also serve as a reference for future me.

The primary reason for this decision is that, based on the evidence currently at my disposal, I believe that the climate impact of in-person attendance is disproportionate with respect to the expected benefits.

A roundtrip flight from Paris to New Orleans emits approximately 2.5t CO2e per traveller, which exceeds single-handedly the 2050 yearly emission target of 2t per capita fixed by the Paris agreements. In other words, intercontinental travel once per year is far from being aligned with the Paris agreements, and has a disproportionate impact on climate. A classical argument in the ML research community is that the target of 2t per capita is an average over the world population, and that research work should get priority over other travel reasons. However, beyond being severely elitist, this argument is a logical fallacy since any traveller could legitimately deem themselves high on the agenda. Is meeting for a week-long research conference really more important than seeing one’s family?

Furthermore, being currently in my twenties, 2050 is not an abstract temporal marker. It is 27 years from now. It is well within the scope of my professional life expectancy, since I will be in my early fifties by then. In 27 years, the ML research community will either have failed to meet the Paris agreements or changed dramatically its current practices. The requirement to change is a very concrete one at the timescale of my career, making it an urgent imperative to think about it as soon as possible and to begin shifting towards more sustainable alternatives.

I measure very well that this decision may have a negative impact on my career, in particular because I stand on the opposite side to the vast majority of the community in declining to attend in person despite being financially and administratively able to. This being said, I intend to take the following steps that will, I hope, mitigate the impact of not presenting my work at NeurIPS in person.

I will remotely attend the main conference.
I will attend a local meetup in Paris.
I am more than willing to present my works at other meetups in Europe. As of now, I have large availability in the weeks before, during, and after NeurIPS.
I will attend either ICLR or ICML in 2024, both taking place in Vienna, which is reachable by train from Paris.

I hope with all my heart that the ML research community will engage in a profound discussion to shift to more sustainable practices. In particular, local meetups seem like the most natural evolution of current worldwide gatherings. They eliminate carbon emissions from long-distance travel. They are accessible to a much wider part of the community, since they are less costly and may be less stringent in terms of visa requirements. And, last but certainly not least, this system has already been shown to work. Recent initiatives of local meetups for NeurIPS in Paris, Zürich, Ghent, Munich or Cambridge, gathering between 150 and 300 researchers, have been unanimously acclaimed by participants and provided a proof-of-concept for these smaller venues.

Of course, such a change in the practices of the community will require a massive effort from many members in the community. If you agree with the diagnosis above, and are willing to help construct together the sustainable research practices of tomorrow, please reach out at pierre.marion@sorbonne-universite.fr. Any constructive feedback can also be addressed to the same email.