A Stein variational Newton method

Gianluca Detommaso, Tiangang Cui, Alessio Spantini, Youssef Marzouk, Robert Scheichl

Research output: Chapter in Book/Report/Conference proceedingConference PaperResearch

3 Citations (Scopus)

Abstract

Stein variational gradient descent (SVGD) was recently proposed as a general purpose nonparametric variational inference algorithm [Liu & Wang, NIPS 2016]: it minimizes the Kullback-Leibler divergence between the target distribution and its approximation by implementing a form of functional gradient descent on a reproducing kernel Hilbert space. In this paper, we accelerate and generalize the SVGD algorithm by including second-order information, thereby approximating a Newton-like iteration in function space. We also show how second-order information can lead to more effective choices of kernel. We observe significant computational gains over the original SVGD algorithm in multiple test cases.

Original languageEnglish
Title of host publicationProceedings of Advances in Neural Information Processing Systems
Subtitle of host publication32nd Conference on Neural Information Processing Systems, NeurIPS 2018; Montreal; Canada; 2 December 2018 through 8 December 2018
Editors S Bengio, H Wallach, H Larochelle, K Grauman, N Cesa-Bianchi, R Garnett
Place of PublicationSan Diego CA USA
Number of pages11
Volume2018-December
Publication statusPublished - 1 Jan 2018
EventAdvances in Neural Information Processing Systems, 2018 - Montreal Convention Center (Palais des Congrès de Montréal), Montreal, Canada
Duration: 2 Dec 20188 Dec 2018
Conference number: 31st
https://nips.cc/Conferences/2018

Publication series

NameAdvances in Neural Information Processing Systems
ISSN (Print)1049-5258

Conference

ConferenceAdvances in Neural Information Processing Systems, 2018
Abbreviated titleNIPS 2018
CountryCanada
CityMontreal
Period2/12/188/12/18
OtherThe Annual Conference on Neural Information Processing Systems (NeurIPS) is a multi-track machine learning and computational neuroscience conference that includes invited talks, demonstrations, symposia and oral and poster presentations of refereed papers. Following the conference, there are workshops which provide a less formal setting.
Internet address

Cite this

Detommaso, G., Cui, T., Spantini, A., Marzouk, Y., & Scheichl, R. (2018). A Stein variational Newton method. In S. Bengio, H. Wallach, H. Larochelle, K. Grauman, N. Cesa-Bianchi, & R. Garnett (Eds.), Proceedings of Advances in Neural Information Processing Systems: 32nd Conference on Neural Information Processing Systems, NeurIPS 2018; Montreal; Canada; 2 December 2018 through 8 December 2018 (Vol. 2018-December). (Advances in Neural Information Processing Systems). San Diego CA USA.