My research focuses on developing statistically principled methods for machine learning systems that operate in uncertain and dynamic environments.
Anomaly Detection
I develop methods for detecting anomalies in streaming data while controlling the rate of false discoveries. This includes online algorithms that adapt to distribution shift and maintain formal statistical guarantees.
Conformal Inference
Conformal prediction provides distribution-free uncertainty quantification for machine learning models. I work on extending these methods to online settings where data arrives sequentially and the underlying distribution may change over time.
Online Learning
I am interested in algorithms that learn and adapt in real-time, particularly methods that provide finite-sample guarantees. This includes online multiple testing procedures and adaptive prediction intervals.
False Discovery Rate Control
Controlling the false discovery rate (FDR) is crucial in many scientific applications. I develop online FDR control methods that handle sequential hypothesis testing with dependent test statistics.