Shan Yang

Multimodal reasoning · post-training (SFT/DPO/RL) · physics-grounded VLMs. Staff Applied Scientist @ Adobe Foundry. PhD UNC (advisor: Ming C. Lin). Prev: Google Research, Amazon. Recent: [Physics-R1](https://arxiv.org/abs/2605.14040) — +18.9 pp on PhysOlym-A via RL post-training. Co-author on Attention Bottlenecks for Multimodal Fusion (NeurIPS 2021), AI Choreographer / AIST++ (ICCV 2021), ICAR (AAAI 2024), VLAP (ECCV 2024).