Home

Statistics · Machine Learning · Climate

Reliable uncertainty for modern machine learning and scientific systems.

My research develops rigorous statistical machine learning methods for uncertainty-aware prediction and decision-making in complex, high-dimensional systems. I focus on calibrated predictive distributions, conformal inference, generative modeling, operator learning, and structured prediction for spatiotemporal data.

Conformal inference Generative modeling Operator learning Climate and environmental systems

Recent work

2026 · NeurIPS, in review

Flow-Based Conformal Predictive Distributions

Turns high-dimensional conformal prediction sets into calibrated distributions that can be sampled and used in downstream scientific tasks.

Paper Code

2026 · Preprint

Manifold Constrained Conformal Prediction for Spatial Events

Builds calibrated, geometrically plausible prediction sets for spatial event clouds such as tropical cyclone genesis and earthquakes.

Paper

2025 · TMLR, in review

Locally Adaptive Conformal Inference for Operator Models

Provides locally adaptive, function-valued uncertainty sets for neural operators with finite-sample statistical guarantees.

Paper Code

Research program

Calibrated generative inference

I develop conformal predictive distributions and flow-based methods that retain finite-sample reliability while representing uncertainty over fields, functions, trajectories, and point processes. The goal is to make modern generative models statistically calibrated and usable for downstream decisions.

Scientific machine learning for climate and spatiotemporal systems

Climate provides a demanding test bed for model evaluation, distribution shift, operator learning, and uncertainty quantification. My work develops distributional metrics for climate-model validation and methods that integrate simulations with observational data to produce more reliable projections.

Dependence, extremes, and environmental decisions

I study spatiotemporal dependence, teleconnections, extremes, and intervention in environmental systems. Applications include compound climate risk, West Nile virus forecasting, mosquito-control policy, and other settings where incomplete monitoring and changing environments complicate prediction.

Conformal climate projections for temperature and precipitation — Conformal ensembles combine climate-model output with observations to produce tighter, calibrated projections.

Map of estimated extremal teleconnection structure — Estimated extremal teleconnection structure reveals regions that may contribute to compound climate risk.

About

I am an Assistant Professor in the Department of Statistics at the University of Connecticut. Before joining UConn in 2024, I was an Assistant Professor of Statistics at Texas A&M University. I received my PhD in Statistics from the University of Illinois Urbana-Champaign. My work has appeared in NeurIPS, the Annals of Applied Statistics, JASA, Technometrics, JABES, and scientific application journals, and has been supported by Sandia National Laboratories and private foundations.