Research

Foundation models are expanding the reach of prediction in business, finance, and science. Tabular models such as TabPFN, TabICL, and TabFM draw on large-scale synthetic pretraining, while time-series models such as TimesFM learn from substantial real-world and synthetic corpora.

Scientific analysis, however, asks for more than prediction. Reliable inference and hypothesis testing are equally important. I am interested in how foundation models can address these statistical tasks while preserving valid uncertainty quantification.

Multimodal AI brings together data types that statisticians have long studied. I am interested in how statistical reasoning about relationships across modalities and uncertainty can make these models more reliable and effective.

As data analysis becomes more automated and accessible to non-statisticians, it becomes increasingly important to evaluate whether the resulting analyses are sound. I am interested in using statistical principles to help automated systems choose appropriate methods, assess their results, and communicate uncertainty.

Genomic studies and electronic health records often involve complex data structures, including multimodal measurements, blockwise missingness, and distribution shift. I am interested in applying principled statistical methods to address these challenges in real-world studies.