Publications

Paul Janson, Edouard Oyallon, Eugene Belilovsky (2026). Learned Subspace Compression for Communication-Efficient Pipeline Parallelism. In ICML 2026 Workshop.

Paul Janson, Edouard Oyallon, Eugene Belilovsky (2026). Stabilizing Native Low-Rank LLM Pretraining. In ICML 2026.

PDF Cite Code Post

Paul Janson, Benjamin Therien, Quentin Anthony, Xialong Huang, Abhinav Moudgil, Eugene Belilovsky (2026). PyLO: Towards Accessible Learned Optimizers in PyTorch. In MLSys 2026.

PDF Cite Code Post

Paul Janson, Vaibhav Singh, Paria Mehrbod, Adam Ibrahim, Irina Rish, Eugene Belilovsky, Benjamin Therien (2025). Beyond Cosine Decay: On the effectiveness of Infinite Learning Rate Schedule for Continual Pre-training. In CoLLAs 2025 (oral).

PDF Cite Code Post

Paul Janson, Tiberiu Popa, Eugene Belilovsky (2024). Towards motion from video diffusion models. In ECCV 2024 Workshop.

Wenxuan Zhang, Paul Janson, Rahaf Aljundi, Mohammed Elhoseiny (2024). Overcoming Generic Knowledge Loss with Selective Parameter Update. In CVPR 2024.

Kai Yi, Paul Janson, Wenxuan Zhang, Mohammed Elhoseiny (2023). Domain Aware Zero shot learning. In ICCV 2023 Workshop.

Wenxuan Zhang, Paul Janson, Divyansh Jha, Kai Yi, Ivan Skorodov, Mohammed Elhoseiny (2023). Continual zero-shot learning through semantically guided generative random walk. In ICCV 2023.

Paul Janson, Wenxuan Zhang, Rahaf Aljundi, Mohammed Elhoseiny (2022). A Simple baseline that questions the use of pre-trained model in continual learning. In NeurIPS2022 Workshop.