Linhai Ma

I am a postdoc at Yale. My current work focuses on clinical NLP and structured generation. In particular, I study methods for turning unstructured domain text into schema-valid records grounded in the source evidence, as well as token-level preference optimization for tasks in which output validity depends on a small number of critical positions. I also work on retrieval and agentic methods in the financial domain, in collaboration with The Fin AI Community.

I received my Ph.D. in Computer Science from the University of Miami under the supervision of Dr. Liang Liang, where I studied the adversarial robustness of deep neural networks. I developed adversarial training methods with sample-specific margins that adapt during training, which improve robustness while largely preserving clean accuracy. I evaluated these methods across several tasks, including ECG diagnosis, MRI segmentation, cephalometric landmark detection, and blood cell detection.

I received my master’s degree from the State Key Laboratory of Computer Science at the Institute of Software, Chinese Academy of Sciences, under the supervision of Dr. Peng Wu. My research focused on adaptive coverage-guided testing of C++ concurrent data structures.

My CV is available here.