Thomas Sesmat
Hi! :)
I’m a Research Scientist at Deezer Research, working on the general understanding of generative models, especially working toward making these crazy things safe and respectful for Arts and Artists. For now, my (AI-oriented) topics of interest are:
- Memorization in generative models
- AI / deepfake detection
My work sits at the intersection of theoretical characterization and empirical observation.
Outside of research, I like building and automating things. Deep learning is one of the tools I reach for, but far from the only one. My very special passion is cooking, which I consider the most complete form of creation… happy to debate that with anyone who disagrees.
If you’re interested in collaborating, or just want to talk through an idea (or whatever comes to mind) feel free to reach out, I’d love to connect! :)
News !
10/08/2026 — New blog post about our ICML paper. It took more time than expected, but I'm very happy with the result :))
20/07/2026 — Update of my previous blog post about diffusion models! I'm now talking about Mean Flow, Shortcut Models and IMM !
01/06/2026 — I joined Deezer Research 💜 as a Research Scientist. How crazy!!
18/04/2026 — I will attend ICML 2026 in Seoul to present our paper "Where Flow Matching Leaks"
28/02/2026 — I am now officially an applied mathematics engineer :))
20/08/2025 — A new blog post is out about the broad family of diffusion-based generative models. There's quite a lot to read, but I believe it's worth the time
01/03/2025 — Just starting new adventure at Telecom Paris x Deezer for my Master Internship 🥳
11/06/2024 — As promised, my new (and first!) blog post, AI in Music: A Historic Review, is out! 🎶
10/25/2024 — Yesterday, I attended a very interesting conference on the intersection of classical music and AI, presented by Jean-François Zygel. Stay tuned for an upcoming post on this topic!
10/20/2024 I was lucky enough to take part in the CIMI Thematique School on high-dimensional machine learning and inference from 10/07 to 10/18!