research

My research focuses on developing reliable machine learning for real-world scientific problems, with an emphasis on quantifying uncertainty, learning from imperfect data, and generalizing under distribution shift.

Real-world scientific data are rarely ideal. In domains such as medical imaging and animal behavior analysis, data can be limited, noisy, imbalanced, or heterogeneous, while models may encounter populations and environments that differ substantially from those seen during training. These challenges motivate my work on machine learning systems that can learn effectively from imperfect supervision, characterize what they do not know, and remain reliable beyond the training distribution.

More broadly, I am interested in how representation learning, particularly in the era of foundation models, can enable such systems to extract meaningful structure from complex scientific data with limited supervision. My long-term goal is to develop learning systems that are not only accurate in controlled settings, but also reliable, data-efficient, and adaptable in the real world.

My current research explores these questions in medical imaging and animal behavior analysis, with broader interests in reliable machine learning for scientific discovery and applications.