CMStatistics 2021: Start Registration
View Submission - CMStatistics
Title: Spike-and-slab variational bayes for high-dimensional survival analysis Authors:  Michael Komodromos - Imperial College London (United Kingdom) [presenting]
Sarah Filippi - Imperial College London (United Kingdom)
Marina Evangelou - Imperial College London (United Kingdom)
Kolyan Ray - Imperial College London (United Kingdom)
Abstract: In recent years variational Bayes (VB) has presented itself as a viable alternative to MCMC, particularly in situations where scalability is key. We follow such developments and present a VB approximation to sparse high-dimensional Bayesian proportional hazards models. Within our VB approximation we utilise a mean-field spike-and-slab variational family, thereby offering mechanisms for variable selection, coefficient estimation and uncertainty quantification. We demonstrate the performance in a variety of simulation settings, as well as demonstrate applicability to real-world datasets.