Léon Zheng

I am a researcher at the Huawei Lagrange Mathematics and Computing Research Center in Paris. I completed my PhD in Computer Science in 2024 at valeo.ai and École normale supérieure de Lyon (LIP, Inria OKCHAM), under the supervision of Rémi Gribonval, Patrick Pérez, Gilles Puy and Elisa Riccietti. I graduated from École Polytechnique (X2016) and Master M2 “Mathématiques, Vision, Apprentissage” at ENS Paris-Saclay in 2020.

My research work focuses on mathematical and practical aspects of resource-efficient deep learning, including:

• Butterfly factorization to enable O(N log N) fast algorithms for matrix multiplication
• Sparse neural networks for inference acceleration on GPU using the butterfly structure
• Self-supervised learning for visual representations without data annotation
• Parallel algorithm for matrix multiplication on NPU architecture
• Diffusion models for image inverse problems (super-resolution, deblurring, inpainting)

News:

• 2026/02: I will participate to the 2026 MBZUAI Machine Learning Winter School on representation learning and generative AI.
• 2025/07: I will participate to the 2025 ICML to present our work on fast GPU inference with Kronecker-sparse matrices.
• 2024/06/28: I will join the Huawei Lagrange Mathematics and Computing Research Center in Paris as a researcher.
• 2024/05/29: I will defend my PhD thesis about “Data frugality and computational efficiency in deep learning” at ENS de Lyon.
• 2023/09: I will present our work on butterfly factorization with unknown permutations at GRETSI 2023.
• 2023/05: I will present our work on butterfly factorization at Université de Mons.
• 2023/01: Our paper on self-supervised learning with rotation-invariant kernels has been accepted at ICLR 2023.
• 2022/11: I will present our work on butterfly matrix factorization during the minisymposium “Butterfly factorizations: Algorithms, applications, and theory” at SIAM CSE’23.