Research

Reliable machine learning for healthcare

I build methods that make clinical predictions more useful, transparent, and dependable under real-world uncertainty.

Research interests

  • Uncertainty quantification and conformal prediction
  • Graph and hypergraph representation learning
  • Clinical NLP and large language models
  • Longitudinal electronic health records
  • Computational epidemiology
  • Trustworthy machine learning for healthcare

Current focus

Distribution-free uncertainty

Developing conformal prediction methods for structured clinical data, graphs, and hypergraphs, with an emphasis on useful prediction sets and reliable coverage.

Representation learning

Learning robust representations from longitudinal EHRs and higher-order relationships among patients, diagnoses, treatments, and clinical events.

Clinical language models

Using large language models to organize clinical notes, reason across specialty-specific perspectives, and improve downstream health-risk prediction.