Research

My research develops theoretical and optimization-driven methods for trustworthy machine learning, with a focus on graph-structured and biomedical data. I study how learning systems can remain fair, robust, privacy-aware, uncertainty-aware, and scalable under real-world deployment constraints. My work spans certifiably robust and fair graph neural networks, fairness-aware knowledge distillation, graph coarsening, graph-language alignment, and trustworthy biomedical representation learning.

Research Areas

Trustworthy and Efficient Graph Representation Learning

I study robustness, fairness, efficiency, and privacy in graph learning. My work includes certifiably robust and fair GNN training under structural and feature perturbations, and GNN-to-MLP knowledge distillation that enables efficient graph-free inference while preserving structural information and fairness. Current work also examines class-conditional transfer disparities and connectivity leakage in graph-free inference.

Graph Coarsening and Scalable Graph Learning

I develop optimization-based approaches for reducing graph size while preserving task-relevant structural and feature information. The goal is to make learning on large graph-structured datasets more computationally efficient without discarding the information needed downstream.

Trustworthy Biomedical Representation Learning

I design self-supervised, robust, and equity-aware learning methods for heterogeneous neuroimaging data. This includes multi-domain fMRI functional-connectivity modeling for major depressive disorder and robust/fair BOLD-signal analysis across population groups and distribution shifts.

Graph-Language and Tabular Large Language Models

I investigate graph-language alignment for neurological data, including mapping self-supervised fMRI graph representations into the embedding space of frozen language models for few-shot diagnosis. I am also studying tabular large language models for healthcare and EHR data across robustness, fairness, privacy, uncertainty, and calibration.

Medical Imaging and Deployable Healthcare AI

My earlier work developed deep-learning methods for CT-based COVID-19 diagnosis in collaborative edge-cloud settings, targeting accurate and computationally practical inference in resource-constrained healthcare environments.

Publications

For a complete citation list, see my Google Scholar profile.

Ongoing Research

• Do MLPs Inherit GNN Knowledge Uniformly? Class-Conditional Transfer Equity in Graph-Free Distillation. Revision submitted to Transactions on Machine Learning Research.
• Coarsened Graph Learning. Under review at Transactions on Machine Learning Research.
• The Hidden Privacy Cost of Graph-Free Inference: Connectivity Leakage in GNN-to-MLP Knowledge Distillation. Manuscript in preparation for ICLR 2027.