Victor Schmidt

Mila - C2084

6666 rue Saint-Urbain

Montréal, QC H2S 3H1, Canada

schmidtv 🌀 mila.quebec

PhD

💡 I am a final year PhD student interested mostly in applications of AI in the fight against climate change. This has led me to explore various areas, from computer vision and GANs to geometric graphs for materials modeling.

1️⃣ My first PhD project focused on creating AI visualizations of what your home could look like if climate-related extreme events (floods, wildfires or smog events) happened there. It’s not about climate projections, it’s about empathy: a climate in which every address in the world is experiencing floods, wildfires and smog at the same time does not exist. But the one we have is getting warmer, more dangerous, and everyone’s actions have global consequences. This is why we published ClimateGAN (ICLR 2022) and deployed it on This Climate Does Not Exist.

2️⃣ The second project I’ve worked on for almost 2 years now is about using AI for materials discovery. The long-term goal is to contribute to the discovery of more efficient electro-catalysts that would improve the energy efficiency of a wide range of chemical reactions. To that end, I have worked on graph neural networks for materials property prediction: PhAST (under review at JMLR) and FAENet (ICML 2023). I am now focusing on using this work within a generative model (GFlowNet) to efficiently explore the huge space of potential electro-catalysts.

3️⃣ Finally, I wanted to keep an eye on my own community and work towards quantifying the carbon emissions of machine learning. This lead to a workshop paper (Climate Change AI workshop, NeurIPS 2019) and its online emissions calculator, and a Python open-source library: codecarbon.

Community work

Beyond publishing, I try to be a good member of our community.

Student Lab Representative for 3 consecutive years at Mila
TA for 2 consecutive years for the UdeM IFT 3710/6759 Advanced projects in Machine Learning course,
I contribute a module on AI & Climate Change to an upcoming MOOC by the Geography Department of Université de Montréal (with the amazing Mélisande Teng) and I like writing tutorials (from as early as 2015 or more recently an introduction to Pytorch).

TL;DR Resume

Prior to this PhD, I was a student at École polytechnique, France (Engineering degree - 2012→2016), and at University College London, UK (MSc. in Machine Learning - 2016→2017). I also worked for about a year as a Public Interest Entrepreneur at the French Ministry for the Econommy and Finance (2018).

Open source

I also love creating open-source libraries. I tend to be the kind of persons who will spend a week or a month automating something that takes a couple hours, and then make an open-source project out ouf it. This is how I came to build PaperMemory a web-based automated reference manager, minydra a minimalist Python argument parser, ntfy-wrapper a simple notification wrapper around the ntfy.sh service, or gitmopy an interactive CLI to make pretty commits following the Gitmoji specification.

In a previous life, I also created metada.org, a website to visualize the hierarchies of ownership in the French media.

selected publications

ClimateGAN: Raising Climate Change Awareness by Generating Images of Floods

Victor Schmidt*, Alexandra Sasha Luccioni*, Mélisande Teng, and 8 more authors

ICLR, 2022

Bib HTML PDF Code

@article{schmidt2021climategan,
  title = {ClimateGAN: Raising Climate Change Awareness by Generating Images of Floods},
  dimensions = {false},
  author = {Schmidt*, Victor and Luccioni*, Alexandra Sasha and Teng, Mélisande and Zhang, Tianyu and Reynaud, Alexia and Raghupathi, Sunand and Cosne, Gautier and Juraver, Adrien and Vardanyan, Vahe and Hernandez-Garcia, Alex and Bengio, Yoshua},
  year = {2022},
  journal = {ICLR},
}

PhAST: Physics-Aware, Scalable, and Task-specific GNNs for Accelerated Catalyst Design

Alexandre Duval*, Victor Schmidt*, Santiago Miret, and 3 more authors

In AI for Accelerated Materials Design NeurIPS 2022 Workshop, 2022

Bib HTML PDF Code

@inproceedings{duval2022phast,
  title = {Ph{AST}: Physics-Aware, Scalable, and Task-specific {GNN}s for Accelerated Catalyst Design},
  dimensions = {false},
  author = {Duval*, Alexandre and Schmidt*, Victor and Miret, Santiago and Bengio, Yoshua and Hern{\'a}ndez-Garc{\'\i}a, Alex and Rolnick, David},
  booktitle = {AI for Accelerated Materials Design NeurIPS 2022 Workshop},
  year = {2022},
  url = {https://openreview.net/forum?id=hHercGKiXvP},
}

FAENet: Frame Averaging Equivariant GNN for Materials Modeling

Alexandre Duval*, Victor Schmidt*, Alex Hernandez Garcia, and 4 more authors

ICML 2023, 2023

Bib HTML PDF Code

@article{duval2023faenet,
  title = {FAENet: Frame Averaging Equivariant GNN for Materials Modeling},
  dimensions = {false},
  author = {Duval*, Alexandre and Schmidt*, Victor and Garcia, Alex Hernandez and Miret, Santiago and Malliaros, Fragkiskos D. and Bengio, Yoshua and Rolnick, David},
  year = {2023},
  journal = {ICML 2023},
}

torchgfn: A PyTorch GFlowNet library

Salem Lahlou, Joseph D. Viviano, and Victor Schmidt

arXiv preprint arXiv: 2305.14594, 2023

Bib HTML PDF Code

@article{lahlou2023torchgfn,
  title = {torchgfn: A PyTorch GFlowNet library},
  dimensions = {false},
  author = {Lahlou, Salem and Viviano, Joseph D. and Schmidt, Victor},
  year = {2023},
  journal = {arXiv preprint arXiv: 2305.14594},
}

Quantifying the Carbon Emissions of Machine Learning

Alexandre Lacoste*, Alexandra Luccioni*, Victor Schmidt*, and 1 more author

Climate Change AI workshop, NeurIPS, 2019

Bib HTML PDF Code

@article{lacoste2019quantifying,
  title = {Quantifying the Carbon Emissions of Machine Learning},
  dimensions = {false},
  author = {Lacoste*, Alexandre and Luccioni*, Alexandra and Schmidt*, Victor and Dandres, Thomas},
  year = {2019},
  journal = {Climate Change AI workshop, NeurIPS},
}