Hi, I’m Yuan YIN 银元
AI Researcher at Valeo.ai
I work on parameter-efficient model adaptation, learning physical dynamics, and autonomous driving. My current research covers perception, motion forecasting, and decision making for self-driving, along with adapting large pretrained models and learning at scale. Across these, I favor methods that exploit the structure of a problem.
I defended my PhD on physics-aware deep learning in June 2023 at Sorbonne Université, supervised by Prof. Patrick Gallinari and Assoc. Prof. Nicolas Baskiotis. I hold a BSc in Computer Science from Beihang University and an MSc from Université Paris Cité and Sorbonne Université.
Research themes
-
Model adaptation
Fine-tuning large pretrained models with few trainable parameters, and making learned models generalize to unseen conditions.
-
Learning physical dynamics
Combining physical models with deep networks, and forecasting PDEs continuously across space and time.
-
Autonomous driving
Motion forecasting, end-to-end planning, and simulation for training and stress-testing driving policies.
Latest updates
Selected publications
- IPA: An Information-Reconstructive Input Projection Framework for Efficient Foundation Model Adaptation
Best Paper at NeurIPS 2025 CCFM Workshop - Physics-Aware Deep Learning and Dynamical Systems: Hybrid Modeling and Generalization
Accessit ex æquo to the AFIA IA 2024 Thesis Prize
