Leo Feng

ML Research Team Lead @ RBC Borealis

prof_pic.jpg

I am a Machine Learning Research Team Lead at RBC Borealis (formerly Borealis AI). Previously, I received my PhD from Montreal Institute for Learning Algorithms and Université de Montréal where I was advised by Prof. Yoshua Bengio working on the design of efficient and scalable deep learning models. Prior to Mila, I did my Bachelors in Computer Science at University of Oxford (WhiRL), working with Prof. Shimon Whiteson and Dr. Luisa Zintgraf on Meta-Learning and Reinforcement Learning.

My research interests include efficient sequence models, long-context LLMs, and evaluating agents and LLMs under uncertainty.

News

Nov 2026 Reviewed for Nature.
Oct 2026 Started as Machine Learning Research Team Lead at RBC Borealis.
Oct 2026 Three papers accepted at NeurIPS 2026 workshops on LLM post-training and agents.
Sep 2026 Two papers accepted at NeurIPS 2026: Can LLMs Take Retrieved Information with a Grain of Salt? and Off-Policy Evaluation of Large Language Models via Learned Semantic Bottleneck Embeddings.
May 2026 Promoted to Senior Machine Learning Researcher at RBC Borealis.

Selected Works

  1. Can LLMs Take Retrieved Information with a Grain of Salt?
    Behzad Shayegh, Mohamed Osama Ahmed, Frederick Tung, and Leo Feng
    In Advances in Neural Information Processing Systems, 2026
  2. Were RNNs All We Needed?
    arXiv, 2024
  3. Memory Efficient Neural Processes via Constant Memory Attention Block
    In International Conference on Machine Learning, 2024
  4. Tree Cross Attention
    In International Conference on Learning Representations, 2024
  5. Latent Bottlenecked Attentive Neural Processes
    In International Conference on Learning Representations, 2023
  6. Towards Better Selective Classification
    In International Conference on Learning Representations, 2023