Our colloquium series offers a diverse platform for research scholars, faculty, students, and industry experts to share and exchange ideas, fostering discussion and networking across mathematics, statistics, and data science.
Professor Hsin-Hsiung "Bill" Huang will speak at this week's colloquium on "Scalable Bayesian uncertainty quantification for mixed membership in multilayer networks via spectral-assisted Gaussian variational inference with degree heterogeneity."
Abstract: Multilayer networks arise when relational data are observed repeatedly across time, conditions, or platforms, such as monthly transportation flows, task-based brain connectivity, or evolving trade networks. A central goal is to infer latent community structure while allowing two common features common in practice: mixed memberships (nodes can belong to multiple communities) and degree heterogeneity (hubs and layer-specific activity shifts). Bayesian modeling provides a principled route to uncertainty quantification, but full posterior computation can be prohibitive at modern network scales, and standard variational approximations can misrepresent posterior dependence and miscalibrate credible sets.
In this talk, Huang will present a scalable Bayesian framework for degree-corrected mixed membership models in multilayer networks. The key device is a spectral-assisted likelihood that takes preliminary spectral estimates of global quantities, yielding a vertex-wise objective that is separable across nodes and enables embarrassingly parallel inference. For each node, Gaussian variational inference is performed using a structured covariance motivated by Fisher information, which captures the essential posterior coupling between membership profiles and degree parameters without the cost of full covariance. He'll also describe posterior theory, including Bernstein–von Mises results for both the exact posterior and the proposed variational approximation, which together imply asymptotically calibrated uncertainty for node memberships. Finally, he'll illustrate the method on U.S. airport transportation networks, comparing pre-COVID and during-COVID regimes and highlighting interpretable shifts in community structure and membership uncertainty.
Speaker Bio: Hsin-Hsiung "Bill" Huang is an associate professor of statistics and data science at UCF and is in the final stage of promotion to full professor (dean-approved; provost approval expected in early April 2026). He earned his doctorate in statistics from the University of Illinois at Chicago and holds master's degrees in statistics (Georgia Tech) and in mathematics (National Taiwan University). His research spans Bayesian modeling and low-rank methods (matrix and tensor modeling), bridging computation and theory to develop scalable statistical methodology for complex, large-scale data, with applications in spatiotemporal modeling, neuroimaging reconstruction, and threat detection. He serves as principal investigator on multiple NSF algorithms for threat detection grants and as a co-investigator on an NIH NINDS R01 and a Lockheed Martin anomaly detection project. In 2025, he was selected as a Fulbright Specialist for an approved visit to Al-Khwarizmi University in Urgench, Uzbekistan, from April to May 2026.
Read More