Yu Gu
顾禹 (Gù Yǔ)
Currently at Microsoft Research and formerly co-founding a Series C startup, I focus on large-scale foundation models, multimodal learning, agentic frameworks, and enterprise AI deployment—turning advanced research into high-impact solutions.
Also known as “Aiden”.
Highlights
- PubMedBERT: Built one of the first domain-adaptive LLMs, achieving 20M+ downloads, 2000+ citations, and ACM HLTH Best Paper of the Year. It now powers Azure’s Text Analytics for Health across 10+ major institutions.
- BiomedParse: Created a universal segmentation model for CT, MRI, and pathology data (Nature Methods), deployed by multiple healthcare organizations and attracting 50K+ monthly downloads.
- BiomedJourney: Developed a text-to-image generation system for dynamic disease progression simulation, addressing scarce-data problems in medical AI.
- Multi-Agent Systems: Demonstrated LLM-driven collaborative reasoning at the World Economic Forum, optimizing tumor board decisions for complex cancer treatments.
- Startup: Co-founded an AI company, scaling from research to enterprise solutions, culminating in a Series C acquisition.
I published in Nature, Cell, ICLR, ACL, and review for NeurIPS, ICML, and Nature journals. My work has been featured by Forbes, CNBC, and showcased at the World Economic Forum.
news
| Feb 20, 2025 | We’re excited to unveil Magma—our flagship multimodal AI project (Multimodal Agentic Model at Microsoft ReseArch). Today, we released Magma on arXiv (2502.13130), along with its Project Page and GitHub repo. The project has already captured significant community attention, with top influencers sharing the news. |
|---|---|
| Feb 02, 2025 | LLaVA-Rad training data—is out, a multi-modal dataset featuring 400,042 X-ray image-text pairs from MIMIC-CXR, enhanced with GPT-4 for accurate report structuring and clarity. |
| Dec 30, 2024 | Satya Nadella shared our multi-agentic MTB project. |
| Dec 20, 2024 | Our #CXRReportGen is featured in Forbes – delivering state-of-the-art performance at half the model size, it is trained on commercially approved data, ensuring adaptability for specialized applications. |
selected publications
-
Magma: A Foundation Model for Multimodal AI Agents2025 -
Domain-specific language model pretraining for biomedical natural language processingACM, 2022 -
Universalner: Targeted distillation from large language models for open named entity recognitionIn ICLR, 2024 -
Biomedjourney: Counterfactual biomedical image generation by instruction-learning from multimodal patient journeysarXiv preprint arXiv:2310.10765, 2024 -
A whole-slide foundation model for digital pathology from real-world dataNature, 2024 -
A foundation model for joint segmentation, detection and recognition of biomedical objects across nine modalitiesNature methods, 2024
