Advancing Functional Data Clustering and Survival Analysis with Variational Inference

Camila P. E. de Souza, PhD

Associate Professor, Department of Statistical and Actuarial Sciences
University of Western Ontario

NOTE: Meet & Greet Camila de Souza from 3-3:30pm in Room 1140

WHEN: Wednesday, November 5, 2025, from 3:30 to 4:30 p.m.
WHERE: Hybrid | 2001 McGill College Avenue, Rm 1140; Zoom
NOTE: Camila de Souza will be presenting in-person at SPGH
 

Abstract

Variational Inference is a method for analytically approximating the posterior distribution in Bayesian models, offering a more computationally efficient alternative to Markov Chain Monte Carlo (MCMC) sampling techniques. In this talk, I will present work from two recent publications, co-authored with my students and collaborators. The first paper applies VI to functional data clustering, where the goal is to identify groups of curves without prior group membership information. Using a B-spline regression mixture model with random intercepts, we developed a novel variational Bayes (VB) algorithm for simultaneous clustering and smoothing of functional data. The second paper focuses on survival data analysis, proposing a VB algorithm for inferring the parameters of the log-logistic accelerated failure time model by incorporating a piecewise approximation technique to address intractable calculations and achieve Bayesian conjugacy. In both papers, we conducted extensive simulation studies to assess the performance of the proposed VB algorithms, comparing them with other methods, including MCMC algorithms. Applications to real data illustrate the practical use of the methodologies. The proposed VB algorithms demonstrate excellent performance in clustering functional data and analyzing survival data while significantly reducing computational costs compared to MCMC methods. The links to the papers are as follows: https://doi.org/10.1007/s11634-024-00590-w and https://doi.org/10.1007/s11222-023-10365-6.

Speaker Bio

Dr. Camila de Souza is an Associate Professor at the Department of Statistical and Actuarial Sciences at the University of Western Ontario. Before joining Western, Dr. de Souza was a postdoctoral fellow at the Shah Lab for Computational Cancer Biology at the BC Cancer Agency Research Centre. She completed her PhD in Statistics at the University of British Columbia (UBC). She is originally from Brazil, where she received her Master’s and Bachelor’s Degrees in Statistics at the University of Campinas. Her research program consists of developing new statistical methods to analyze large and complex data structures arising from various areas in the Natural Sciences, Health, and Engineering. Dr. de Souza conducts research on techniques involving clustering, hierarchical mixture models, mixed-effects models, hidden Markov models, non-parametric regression, semi-parametric models, the expectation-maximization (EM) algorithm, and Bayesian variational inference. Her website is https://www.desouzacpe.com/