Carbon Impact
(Webpage in progress, comments welcome. Last updated September 1st, 2025.)
Overview
To limit the impact of climate change, it is critical to curb CO2 emissions. Machine learning (ML) research is an important source of emissions through two main causes: training models and air travel to conferences/lab visits. There are also two approaches to curbing emissions, an individual level (''Everything else being the same, what can I do to reduce my own emissions?'') and a collective level (''Can I contribute to building tools that enable everyone to lower their emissions?''). This makes in total 2x2=4 possible directions given in the following table.
| Emission source | Individual level | Collective level |
|---|---|---|
| Training models | Adopting best practices when training models | Work on research questions related to sustainability (energy efficiency of ML models, using ML to reduce emissions in other sectors, etc.) |
| Air travel | Fly less | Build viable alternatives to frequent intercontinental travel for the research community |
The remainder of this page gives details on my efforts, which have focused so far on the air travel side.
Why am I writing this? First for my own record keeping. Towards the end of my PhD, I was considering quitting research, and one of the reasons was to transition to an occupation related to the climate crisis. After much thought, I figured that I would be more happy continuing research, but I pledged to myself to work towards sustainability of my own professional activity as well as of the field in general. For this reason, I want to keep track of where I stand. Besides, it might prove useful to other researchers who are questioning on the carbon impact of their activity and/or want to pick up some ideas.
Why am I focusing so much on air travel? Because I see a clear path forward in this direction. If the opportunity arises in the future, I would also be keen on working on sustainability-promoting ML projects.
My efforts regarding air travel focus on the two directions given in the table above: limit my own air travels, and build viable alternatives to frequent intercontinental travel for the research community. I details those two ideas next.
Limit my own air travels
Air travel represent one of the main areas of my personal climate emissions (as is the case for many priviledged people). This makes a strong case for reducing my travels.
Inspired by Marc-Olivier Renou, I list below my air travels (both personal and professional) since I started my PhD in September 2020. Note that this only represent a part of the travel footprint (e.g., it does not include the cost of staying in hotels).
Total since 2020: 3366kg, decomposed as follows:
- 2025, New York-Paris, one-way (5857km, 778 kg CO2e)
- 2025, Salt Lake City-New York, one-way (3160km, 445 kg CO2e)
- 2025, San Francisco-Salt Lake City, one-way (964 km, 248 kg CO2e)
- 2025, Las Vegas-San Francisco, one-way (670 km, 171 kg CO2e)
- 2025, Vancouver-Las Vegas, one-way (1597km, 271 kg CO2e)
- 2025, Berkeley-Vancouver, one-way (128Okm, 263 kg CO2e)
- 2025, Paris-Berkeley, one-way (8962km, 1190 kg CO2e)
- 2023, Paris-Berkeley, roundtrip (8962km, 2380 kg CO2e)