I'm a second-year Ph.D. student in Computer and Information Science at the University of California, Davis. The pronunciation of my name is “Ray Tsai” (you can also call me Yanson, 言燊).

My research focuses on improving the reliability of multimodal large language models (MLLMs). I work on understanding why multimodal models fail, how humans evaluate multimodal outputs, and how to align models to behave more robustly and predictably.

More specifically, my interests include:

• Multimodal Human Preference Alignment — understanding how humans evaluate multimodal outputs and aligning models with nuanced cross-modal preferences.
• Robustness on Multimodal Alignment — diagnosing and mitigating failure modes in multimodal alignment stage.
• Visual Reasoning & Evaluation — improving MLLMs' ability to understand, reason over, and generate visual information, and designing evaluation methods that capture the limits of these capabilities.

I am an enthusiastic fan of Synthetic Ascension¹.

Email : [first name][last name] [at] [my phd school] [dot] edu