Photo of Rifat Arefin

Rifat Arefin

Ph.D. Candidate
Mila & Université de Montréal
rifat.arefin515@gmail.com

On the job market for research scientist roles




About

I’m a final-year Ph.D. candidate at Mila and Université de Montréal, advised by Irina Rish. I study how large language models learn: how the optimizer, the architecture and the training data shape what a model ends up representing, and how to use that understanding to train better models. I usually start from a precise question, answer it with theory where I can, and then test the answer on real language models.

I’ve done research internships at Meta (twice), Amazon, Samsung and ServiceNow, and have worked with Yoshua Bengio and Yann LeCun.

What I’m working on:

News

Earlier news
  • Sep 2026TT-Delta (KV-cache compression) accepted at the Efficient Reasoning Workshop @ COLM 2026.
  • Jun 2026Started a second research internship at Meta.
  • Jan 2026Finished my research internship at Meta.
  • Aug 2025Started a research internship at Meta.
  • May 2025Layer by Layer accepted at ICML 2025 as a spotlight.
  • Apr 2025Presented Seq-VCR at ICLR 2025.
  • Feb 2025Started a research internship at Samsung AI Lab, Montreal.
  • Jan 2025Seq-VCR accepted at ICLR 2025.
  • Oct 2024Started an Applied Scientist internship at Amazon, Seattle.
  • Oct 2024Paper on representation compression in LLMs accepted at the Compression Workshop @ NeurIPS 2024.
  • Jun 2024Paper accepted at the Foundation Models in the Wild Workshop @ ICML 2024.
  • May 2024Joined ServiceNow Research as a visiting researcher.
  • May 2024CoBalT, on mitigating spurious correlations, accepted at ICML 2024.
  • Jan 2024Passed my Ph.D. candidacy exam.
  • Jun 2023Started an internship at Recursion Pharmaceuticals.

Selected publications

All publications (13)

2026

2025

2024

2022

2021

2020

Industry experience

Service and awards

Reviewer: NeurIPS 2024, 2025, 2026; ICML 2025, 2026; ICLR 2025, 2026, 2027; AISTATS 2024