Conference proceeding
WASP: Scalable Bayes via barycenters of subset posteriors
Proceedings of the Eighteenth International Conference on Artificial Intelligence and Statistics, Vol.38, pp.912-920
01/27/2015
Abstract
The promise of Bayesian methods for big data sets has not fully been realized due to the lack of scalable computational algorithms. For massive data, it is necessary to store and process subsets on different machines in a distributed manner. We propose a simple, general, and highly efficient approach, which first runs a posterior sampling algorithm in parallel on different machines for subsets of a large data set. To combine these subset posteriors, we calculate the Wasserstein barycenter via a highly efficient linear program. The resulting estimate for the Wasserstein posterior (WASP) has an atomic form, facilitating straightforward estimation of posterior summaries of functionals of interest. The WASP approach allows posterior sampling algorithms for smaller data sets to be trivially scaled to huge data. We provide theoretical justification in terms of posterior consistency and algorithm efficiency. Examples are provided in complex settings including Gaussian process regression and nonparametric Bayes mixture models.
Details
- Title: Subtitle
- WASP: Scalable Bayes via barycenters of subset posteriors
- Creators
- Sanvesh SrivastavaVolkan CevherQuoc Tran DinhDavid B Dunson
- Resource Type
- Conference proceeding
- Publication Details
- Proceedings of the Eighteenth International Conference on Artificial Intelligence and Statistics, Vol.38, pp.912-920
- Language
- English
- Date published
- 01/27/2015
- Academic Unit
- Statistics and Actuarial Science
- Record Identifier
- 9984257734802771
Metrics
3 Record Views