Multi-fidelity GNNs for Flood Hazard Mapping
A coarse-to-fine graph learning framework for high-resolution flood hazard mapping on unstructured meshes.
Research and Selected Projects
This page highlights both published and ongoing research projects across scientific machine learning, flood forecasting, power systems, transportation resilience, uncertainty quantification, and large-scale infrastructure analytics. Each summary is written to make the problem, technical contribution, and practical relevance of the work easy to understand.Geometry-Informed Neural Operators for Flood Forecasting
A transferable flood forecasting framework that combines geometry-informed neural operators with domain adaptation for rapid prediction on new river segments.
HydroGraphNet and NVIDIA PhysicsNeMo
A physics-informed and interpretable flood forecasting model that improves accuracy, mass conservation, and operational reliability on unstructured meshes.
Multi-fidelity GNN Surrogates for Probabilistic Power Flow
A graph-based surrogate modeling approach for repeated probabilistic power flow analysis under demand and renewable uncertainty.
Connected-Vehicle Data for Proactive Safety Analytics
A multimodal safety analytics pipeline built on more than 15.6 billion connected-vehicle events to support infrastructure and enforcement decisions.
Graph-based Diffusion for Constrained Optimization
Ongoing postdoctoral work on diffusion models, differentiable optimization, and feasibility-aware decision-making for resilient power and infrastructure systems.
Transportation Resilience for Emergency Medical Response
A probabilistic framework for evaluating how transportation disruption affects emergency medical response and social loss after earthquakes.
Transportation Risk and Recovery for Community Resilience
A probabilistic recovery framework for quantifying transportation-system risk, loss, and resilience at the community scale.
Optimal Design of Experiments for Surrogate Modeling
A backward-greedy sampling strategy for improving the efficiency and accuracy of polynomial-chaos surrogate construction.
End-to-End GNNs for Stormwater Systems
A spatiotemporal graph neural network surrogate for rainfall-driven hydraulic prediction in urban stormwater networks.
Stormwater Digital Twins and Constrained Real-Time Control
Ongoing postdoctoral research on uncertainty-aware flood-state reconstruction and projected diffusion policies for stormwater real-time control.
Flood Sensing, Bayesian Inference, and Sensor Design
Ongoing postdoctoral research on physics-consistent flood-state inference and value-of-information sensor network design.
Neural-Operator Domain Decomposition for Flood Forecasting
Ongoing work on local-to-global neural-operator surrogates for fast shallow-water flood forecasting and uncertainty quantification.