Physics-R1: An Audited Olympiad Corpus and Recipe for Visual Physics Reasoning
Paper • 2605.14040 • Published • 2
Shan Yang
shanyangmie
·
AI & ML interests
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).
Recent Activity
updated a model about 4 hours ago
shanyangmie/physics-r1-seed17 published a model about 4 hours ago
shanyangmie/physics-r1-seed17 updated a model about 4 hours ago
shanyangmie/physics-r1-seed23Organizations
None yet
authored a
paper about 13 hours ago
submitted a
paper to Daily Papers 1 day ago
authored 3
papers 1 day ago
VLAP: Efficient Video-Language Alignment via Frame Prompting and Distilling for Video Question Answering
Paper • 2312.08367 • Published • 1
Attention Bottlenecks for Multimodal Fusion
Paper • 2107.00135 • Published • 1
Beyond Test-Time Training: Learning to Reason via Hardware-Efficient Optimal Control
Paper • 2603.09221 • Published