Scalable Spike-and-Slab
Abstract
Spike-and-slab priors are commonly used for Bayesian variable selection, due to their interpretability and favorable statistical properties. However, existing samplers for spike-and-slab posteriors incur prohibitive computational costs when the number of variables is large. In this article, we propose Scalable Spike-and-Slab (), a scalable Gibbs sampling implementation for high-dimensional Bayesian regression with the continuous spike-and-slab prior of George and McCulloch (1993). For a dataset with observations and covariates, has order computational cost at iteration where never exceeds the number of covariates switching spike-and-slab states between iterations and of the Markov chain. This improves upon the order per-iteration cost of state-of-the-art implementations as, typically, is substantially smaller than . We apply on synthetic and real-world datasets, demonstrating orders of magnitude speed-ups over existing exact samplers and significant gains in inferential quality over approximate samplers with comparable cost.
Keywords
Cite
@article{arxiv.2204.01668,
title = {Scalable Spike-and-Slab},
author = {Niloy Biswas and Lester Mackey and Xiao-Li Meng},
journal= {arXiv preprint arXiv:2204.01668},
year = {2022}
}
Comments
Accepted to ICML 2022. Open-source software in Python and R available at https://github.com/niloyb/ScaleSpikeSlab