Methodology for complex scientific data
My research develops statistical methodology for longitudinal, high-dimensional, network-structured, biomedical, causal, and nonlinear data. Across projects, the goal is to combine rigorous inference, scalable computation, reproducible software, and scientifically meaningful applications.
Longitudinal & Clinical Trials
Rank-based inference, multivariate outcomes, missing data, design, power, and robust treatment-effect analysis.
Imaging & Neurodegeneration
Alzheimer’s, Parkinson’s, CT/MRI, survival, digital twins, and virtual-trial methodology.
Nonparametric & Robust Inference
U-statistics, rank methods, global testing, resampling, and distribution-free procedures.
Data Science & AI
Machine learning, graph learning, representation learning, AI/LLMs, misinformation, and structured high-dimensional systems.
Time Series & Higher-Order Dependence
Polyspectra, nonlinear processes, quadratic prediction, and higher-order frequency-domain inference.
Causal Inference & Discovery
Causal discovery, IV/MR, network-aware estimation, graph-structured causal learning, and biomedical causal analysis.
Longitudinal & Clinical Trial Statistics
A central research program develops nonparametric and robust procedures for multivariate longitudinal outcomes. This includes the Longitudinal Rank-Sum Test (LRST) framework and related methods for multiple endpoints, multiple treatment arms, incomplete outcomes, and covariate-adjusted analysis.
Current directions include:
- global rank-based tests for multivariate longitudinal treatment effects;
- multi-arm and multi-endpoint clinical trials;
- power and sample-size calculation;
- missing-data methods for incomplete longitudinal trials;
- covariate adjustment using regression and machine-learning residualization;
- observational-study extensions using weighting and doubly robust ideas;
- goodness-of-fit procedures for longitudinal covariance structures.
Representative work: Journal of Multivariate Analysis (2025); Statistics in Medicine (2025); Statistics in Biopharmaceutical Research (2025); ongoing work on incomplete-data and residual LRST methods.
Imaging & Neurodegenerative Disease
I collaborate on statistical and computational methods for Alzheimer’s disease, Parkinson’s disease, CT imaging, digital twins, and virtual clinical trials. These projects use survival analysis, longitudinal biomarkers, imaging-derived distributions, graph-based features, and multimodal information.
- survival prediction of cognitive decline using longitudinal biomarkers;
- imaging-based causal analysis and Mendelian randomization;
- MRI/CT feature extraction and distribution-valued predictors;
- disease staging and progression in Parkinson’s disease;
- virtual patient populations and digital-twin methodology;
- imaging anomaly detection and computational validation.
Nonparametric & Robust Inference
A recurring theme throughout my work is inference that remains reliable when standard parametric assumptions are strained. I develop rank-based procedures, U-statistics, global tests, resampling methods, and distribution-free or semiparametric approaches for structured data.
This program connects longitudinal testing, nonlinear time series, socioeconomic information fusion, and high-dimensional structured problems. The common goal is to construct inferential procedures with transparent targets and robust operating characteristics.
Data Science & AI
This program develops and applies modern computational methods for complex structured data. It includes machine learning, graph-based learning, representation learning, AI-assisted scientific workflows, and statistical analysis of high-dimensional network and population data.
- graph neural networks and representation learning;
- AI-generated misinformation and cross-prompt generalization;
- adversarial and robustness evaluation of AI systems;
- graph learning for network-structured data;
- spatial, environmental, and population data science;
- AI/LLM-assisted statistical workflows and research tools.
Nonlinear Time Series & Higher-Order Dependence
My time-series research studies dependence that cannot be fully characterized by autocovariances or second-order spectra. I work on polyspectra, polyspectral means, higher-order frequency-domain analysis, nonlinear prediction, and resampling.
- polyspectral mean estimation for general nonlinear processes;
- polyspectral factorization;
- quadratic prediction using auto-cumulants;
- subsampling and convolved-bootstrap inference;
- higher-order dependence measures for financial and public-health time series.
Representative work: Bernoulli (2026); Annals of the Institute of Statistical Mathematics (2026); Sankhya A (2024).
Causal Inference & Discovery
My causal-methodology work addresses settings in which candidate causes, exposures, biomarkers, or outcomes are high-dimensional and structurally related. The program combines causal inference, graphical models, instrumental variables, Mendelian randomization, and graph regularization.
Current directions include:
- Penalized FCI for sparse causal structure learning;
- network-aware instrumental-variable regression;
- graph-fused regularization for causal node discovery;
- Mendelian randomization for imaging and biomarker systems;
- latent/network-aware causal discovery;
- causal inference for longitudinal and biomedical data;
- causal representation learning for complex multimodal data.