Stat Cafe - Srijato Bhattacharyya
Consistent Spatial Clustering of Complex Objects via Bregman Geometry
- Time: Tuesday, 11/17/2026 from 11:10AM to 12:25PM
- Location: BLOC 448
Description
Spatial clustering of complex-object data presents challenges when conventional parametric likelihoods are difficult to specify. We develop a generalized Bayesian framework for identifying spatially contiguous clusters of complex objects, including probability distributions, networks, and structured matrices. The proposed method uses Bregman divergences to measure within-cluster heterogeneity, geometry-matched priors for tractable marginalization, and a spanning-tree partition prior to enforce spatial contiguity. We establish posterior consistency for spatial partitions and cluster representatives under infill-domain asymptotics. Simulation studies demonstrate accurate recovery of spatial clusters, while an application to racial-composition data from Houston identifies geographically coherent clusters with distinct demographic compositions.
Our Speaker
Srijato Bhattacharyya is a fifth-year Ph.D. student in Statistics advised by Dr. Huiyan Sang and Dr. Bani Mallick. Before joining Texas A&M, he received an M.Stat. from the Indian Statistical Institute and a B.Sc. in Statistics from Presidency University, Kolkata. His research focuses on Bayesian spatial modeling, with applications in demography and public health.