Research

I develop theoretically grounded and computationally efficient methods at the intersection of generative modeling, graph machine learning, signal processing, and optimization, with applications in complex industrial systems.

Generative Modeling for Structured and Dynamical Data

I study how dependencies across variables and time can inform generative modeling. I develop flow matching methods that incorporate graph structure into source distributions and probability paths, connecting generative learning with graph signal processing. These works were developed in collaboration with Jhony H. Giraldo, Aref Einizade, Wenbin Wang, and Olga Fink.

Overview of GRASP: graph-spectral flow matching and velocity-based anomaly scoring, compared with a linear probability path.

Main publications:


Robust and Generalizable Graph Neural Networks

I develop graph neural networks that adapt to noisy graphs and distribution shifts. Using optimization and algorithm unrolling, I design message passing methods that jointly learn node representations and graph structures to improve robustness and generalization across domains. These works were developed in collaboration with Ziping Zhao, Zengfeng Huang, Songtao Lu, and Olga Fink.

Adaptive-structure message passing jointly updates graph signals and graph structure to produce denoised outputs.

Main publications:


Optimization and Algorithm Unrolling

I study the connections among minorization-maximization, fractional programming, and gradient-based methods to develop efficient algorithms for nonconvex optimization. I also develop unrolled networks that learn adaptive updates for beamforming and resource allocation. These works were developed in collaboration with ZheXian Yang, Ziping Zhao, Kaiming Shen, Daniel P. Palomar, and Wei Yu.

MM-Net recurrent unfolding architecture with learned adaptive majorization parameters and projected beamforming updates.

Main publications: