About Me

  • I am a final-year PhD student at Mila.
  • My research broadly focuses on understanding the principles behind how and why deep learning models work, in particular from a data-driven perspective with a current focus on sequence models. As of late, I focus on various foundation model paradigms (attention, linear models, diffusion, tabular models) and how they manipulate information in order to generate faithful and trustworthy responses that can be statistically validated. Additionally, I am intested in more grounded evaluation of such models through proper uncertainty quantification and calibration.
  • On a more applied side, I also dabble in front-facing and system-level research, such as agentic systems and efficient kernel-level programming.
  • I am grateful to have been supported by NSERC, FRQNT, Hydro-Québec, 日本学術振興会 and various institutional scholarships during my studies, allowing me to conduct research on topics of personal fulfillment and interest.
  • I anticipate graduating in the 2026-2027 academic year and am looking for industrial research scientist (or post-doctoral) positions; if you see a fit within your group, please do not hesitate to contact me. References are available upon request.

Relevant Works

Some of my most representative works are listed below. For a more complete list, refer to DBLP or OpenReview.
  • Attention with Routed-Memory for Learnable Sparse Control [ICML 2026]
  • Mamba Modulation: On the Length Generalization of Mamba Models [NeurIPS 2025]
  • Resona: Improving Context Copying in Linear Recurrence Models with Retrieval [COLM 2025]
  • Calibrated Language Models and How to Find Them with Label Smoothing [ICML 2025]
  • ZETA: Leveraging Z-order Curves for Efficient Top-K Attention [ICLR 2025]
  • How Well Can a Long Sequence Model Model Long Sequences? [COLING 2025]
  • Context-Aware Assistant Selection for Improved Inference Acceleration with Large Language Models [EMNLP 2024]

Contacting Me

  • I am always reachable through e-mail, with the addresses structured as "[first_name].[last_name]@[domain_name]" for both Mila (mila.quebec) and Université de Montréal (umontreal.ca). Though I try to be as responsive as possible, do anticipate potential delays of up to 48 hours in case I am busy.