News
- A special recognition for our work: “Across the Loss Landscape with Progressive Growth” received the Best Paper Award at ECML-PKDD 2026!
- I am now Associate Professor at LAMSADE, Université Paris Dauphine – PSL, in the “MILES Team” .
- Next stop, Budapest! “Structured-Sparse Attention for Entity Tracking with Subquadratic Sequence Complexity” has been accepted to the main conference at EMNLP 2026.
- New support for my research: I have been awarded a Chaire Tremplin grant from PRAIRIE-PSAI to support my research over the next three years.
Research
My research focuses on making deep learning more computationally efficient by rethinking how neural networks learn and process information. I am particularly interested in efficient learning algorithms and neural architectures, and in how gradients and information propagate through deep networks.
I investigate how deep networks can be trained more efficiently, how attention mechanisms can handle long sequences at lower computational cost, and how architectures can adapt their capacity during learning. More broadly, I explore how models can learn effectively from limited data and supervision. These questions connect my work on language models and scientific machine learning, with an emphasis on understanding the trade-offs between efficiency, generalization, and robustness.
From 2019 to 2023, I was a PhD student at LORIA under the supervision of Christophe Cerisara. I then joined LAMSADE as a postdoctoral researcher from 2023 to 2025, working with Alexandre Allauzen. From 2025 to 2026, I was a PSL AI Fellow. Since the 1st of September 2026, I am an Associate Professor at Université Paris Dauphine – PSL.
I also enjoy working with PhD students and research interns. Current and former students I have supervised or co-supervised include:
- Hangyue Zhao — former intern, now a PhD student co-supervised with Alexandre Allauzen.
- Alex Colagrande — former intern, now a PhD student co-supervised with Alexandre Allauzen.
- Erwan Fagnou — PhD student, co-supervised with Alexandre Allauzen.
- Joey David — research intern.
- Nan An — former intern.
- Zakariae Moutaouakil — former intern, co-supervised with Blaise Delattre.
Interested in joining?
I welcome enquiries about research internships and PhD opportunities in frugal deep learning. If you are interested in working together, please get in touch with your CV and a short description of your research interests.
Publications
-
Structured-Sparse Attention for Entity Tracking with Subquadratic Sequence Complexity — EMNLP 2026 · Main
-
Across the Loss Landscape with Progressive Growth — ECML PKDD 2026 · Best Paper Award 🏆
-
Trading Complexity for Expressivity Through Structured Generalized Linear Token Mixing — ICML 2026
-
Certified Robustness under Heterogeneous Perturbations via Hybrid Randomized Smoothing — ICML 2026
-
Scaling Direct Feedback Learning with Jacobian Alignment Guarantees — ICLR 2026
-
Limits of Resolution Equivariance in Fourier Neural Operators — AI&PDE · ICLR 2026 Workshop
-
Forward Only Learning for Orthogonal Neural Networks of Any Depth — ECAI 2025
-
Linear Attention with Global Context: A Multipole Attention Mechanism for Vision and Physics — ICCV Workshops 2025
-
Bridging the Theoretical Gap in Randomized Smoothing — AISTATS 2025
-
Accelerated Training through Iterative Gradient Propagation Along the Residual Path — ICLR 2025 · Oral
-
Chain and Causal Attention for Efficient Entity Tracking — EMNLP 2024
* Equal contribution.
For older publications and my full research record, see Google Scholar .