Samuel Lavoie

• Modeling caption diversity in contrastive vision‑language pre‑training. ICML 2024.
• Sparo: Selective attention for robust and compositional transformer encodings for vision. ECCV 2024.
• Language model alignment with elastic reset. NeurIPS 2024.
• Improving compositional generalization using iterated learning and simplicial embeddings. NeurIPS 2024.
• Simplicial Embeddings in Self‑Supervised Learning and Downstream Classification. ICLR 2022 (Oral).
• Integrating Categorical Semantics into Unsupervised Domain Translation. ICLR 2020.
• Learning deep representations by mutual information estimation and maximization. ICLR 2018.

Full list on Google Scholar.

• Compositional Discrete Latent Code for High‑Fidelity, Productive Diffusion Models. In submission at NeurIPS 2025.
• An introduction to Visual Language Modeling. arXiv:2405.17247.
• A surprisingly simple technique to control the pre‑training bias for better transfer: Expand or Narrow your representation. arXiv:2405.17247.

• Lecturer, Université de Montréal — 2025.
• Visiting Researcher, Meta AI Research (FAIR) — 2022‑2024.
• Head Teaching Assistant, Université de Montréal — 2022, 2025.
• Teaching Assistant, Université de Montréal — 2019, 2020.
• Research Intern, Université de Montréal — 2016‑2017.
• Software Engineering Intern, Microsoft — Summer 2016.

• Oral presentation at ICLR & NeurIPS SSL workshop (2022).
• Talk at Samsung Workshop (2022).
• Reviewer for NeurIPS, ICML, and ICLR (2021‑present).
• Organizer, CS Games AI Competition (2019).