Our group develops methods for decision-making in the real world.

A “real-world” method is compatible with the world as it is, not with some idealized notion of it. Real-world methods are

  • necessary to solve practical problems

  • valid without making unrealistic assumptions

  • accessible to users

The tools we build are generally categorized in the fields of nonparametric statistics, causal inference, and machine learning. Our problem areas include risk modeling with electronic health records, randomized trial design, and comparative effectiveness studies.


Fast-converging, scalable supervised learning.

Easy semiparametric probabilistic regression.

Unbiased use of old data in randomized experiments.