Learning when events happen — even when event times are partially observed.

Conformal inference for survival prediction

Distribution-free prediction sets and calibrated uncertainty for complex survival data, including settings where calibration outcomes are partially observed.

Deep Gaussian process AFT models

Flexible, individualized predictive distributions for time-to-event outcomes under an accelerated-failure-time structure.

Applications · ADNI · TCGA

Semi-supervised learning for survival data

Leveraging partially labeled and unlabeled data to improve estimation and prediction when survival outcomes are incompletely observed.

Regression with right-censored covariates

Estimation and valid inference when a predictor — rather than the outcome — is right-censored or otherwise coarsened.

Causal inference with coarsened treatment timing

Estimating treatment and policy effects when treatment timing or adoption time is observed only through sparse intervals.

Policy effects on recurrent-event rates

Methods for evaluating policy changes when outcomes can recur and policy adoption varies across units and time.

Application · NCANDS Child maltreatment

Mediation for survival outcomes

Decomposing survival differences into direct and indirect pathways under censoring.

Application · K-CURE Breast cancer cohort

Gene-panel selection in high dimensions

Selecting small, interpretable gene panels for disease and survival outcomes when candidate features vastly exceed the sample size.

Applications · Alzheimer's disease · cancer genomics

Combinatorial optimization for variable selection

Studying subset-selection solutions produced by mathematical optimization, simulated annealing, and quantum annealing, with emphasis on near-optimal solution landscapes and scientific interpretability.