Shih-Yao (Mike) Lin

Shih-Yao (Mike) Lin, Ph.D.

LLM Post-Training · Agentic AI · Multimodal Intelligence

About Me

I'm a Senior Applied Scientist at Microsoft CoreAI, focusing on LLM post-training, Reinforcement fine-tuning, and Agentic AI optimization. My work bridges model-level learning and agent-level evaluation, with an emphasis on improving the capabilities, reliability, and efficiency of AI systems.

Previously, I was a Senior Applied Scientist at Amazon, working on multimodal foundation model post-training, video understanding, and efficient reasoning. Earlier, I held research positions at Sony, Tencent America, and Bosch, focusing on computer vision and multimodal intelligence.

My research interests span LLM post-training, reinforcement learning, agentic AI, and multimodal intelligence, with the broader goal of building AI systems that can perceive efficiently, reason effectively, and act reliably.

Mount Rainier — a reminder to keep perspective and aim higher.

News

2026-10: 1 paper accepted to WACV'27
2026-10: A Microsoft Foundry Blog: "Information Extraction with Low-Inference-Cost Models" — achieving an 85% F1 gain with Microsoft Foundry Interactive Training.
2026-09: New Microsoft Foundry Blog: “Your Agent Found the Right Schema, Then Ignored It.” — improving schema adherence and reliability in AI agents.
2026-08: New Microsoft Foundry Blog: When Does RL Actually Help Fine-Tuning? — a difficulty-controlled study of SFT vs. reinforcement fine-tuning.
2026-03: Joined Microsoft as a Sr. Applied Scientist

2026-03: 1 paper accepted to ICPR'26
2026-01-07: 1 paper accepted to WACV'26 Workshop
2022-08-08: Join Amazon as an Applied Scientist
2022-04-22: 1 paper accepted to IJCAI'22
2021-10-15: 1 paper accepted to BMVC'21
2021-09: Join Sony as a Sr. Research Scientist
2020-11-02: 2 papers accepted to WACV'21
2020-07-25: 1 paper accepted to ACM Multimedia'20
2019-12-08: Our Artificial intelligence apps for Parkinson's patient evaluation was reported by BBC.
2019-09-16: 1 paper was accepted by WACV'20 (First Round Early Accepted)
2019-07-01: 1 paper was accepted by BMVC'19
2019-05-09: We launched our Parkinson’s AI clinical trial in London, UK, and it was reported by Parkinson’s Life
2019-05-06: Our Parkinson’s AI diagnosis program was reported by Financial Times [PDF]
2019-01-21: Our work entitled "A Pilot Study to Evaluate the Severity of Motor Dysfunction in Parkinson's Disease Based on AI Video Analysis" has been registered in US NIH-registered Clinical Trial
2018-12-11: Our PD evaluation system was demonstrated at T-Day exhibition and it was reported by Sina and Weixin (Chinese)
2018-12-08: Our AI-Assisted PD motion assessment system was demonstrated at PDMD'18 in Hong Kong and it was reported by Synced and HealthcareWeakly.
2018-10-08: Our Parkinson's patient evaluation and grading system was reported by BBC, Bloomberg, Forbes.
2018-09-03: One paper was accepted by BMVC'18

Page updated

Google Sites

Report abuse