Paris, Île-de-France, France
*Please reach out by email* My research interests lie around hardware-friendly AI, including (but not restricted to) : physical computing (classical & quantum), hardware-friendly learning algorithms, temporal credit assignment, quantization algorithms, I/O-aware algorithms & models, distributed training, AI-automated logic design. I am always happy to chat around these topics or provide more general career advice, so don't hesitate to reach out directly on my personal email address [email protected].
Working on AI-assisted AI hardware design.
Co-developing algorithms and analog/digital hardware for scalable neural networks inference and training. - Main achievements: provided engineering support across three different teams, filed two patents as primary author, co-authored four research publications (NeurIPS 2023, NeurIPS 2024, main conference and workshops), won one best workshop paper award, built an internal libraries for quantization-aware training and second-order optimization, built an academic partnership with Mila, lead multiple academic collaborations. - Fields covered: quantization algorithms, second-order optimization, bilevel optimization, energy-based learning, implicit models.
IBM Research Paris-Saclay, working on AI safety. Fields covered: uncertainty quantification, out-of-distribution detection, model calibration, object detection. Mostly product-oriented (unpublished) research.
Working (remotely) under the supervision of Yoshua Bengio and Blake Richards on biologically plausible deep learning. https://arxiv.org/abs/2201.13415 (accepted to ICML 2022)
See description in the education section.