A Stein variational Newton method

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

A Stein variational Newton method

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

School of Mathematics

Research output: Chapter in Book/Report/Conference proceeding › Conference Paper › Research

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 language	English
Title of host publication	Proceedings of Advances in Neural Information Processing Systems
Subtitle of host publication	32nd 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 Publication	San Diego CA USA
Number of pages	11
Volume	2018-December
Publication status	Published - 1 Jan 2018
Event	Advances in Neural Information Processing Systems, 2018 - Montreal Convention Center (Palais des Congrès de Montréal), Montreal, Canada Duration: 2 Dec 2018 → 8 Dec 2018 Conference number: 31st https://nips.cc/Conferences/2018

Publication series

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

Conference

Conference	Advances in Neural Information Processing Systems, 2018
Abbreviated title	NIPS 2018
Country	Canada
City	Montreal
Period	2/12/18 → 8/12/18
Other	The 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	https://nips.cc/Conferences/2018

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.

Detommaso, Gianluca ; Cui, Tiangang ; Spantini, Alessio ; Marzouk, Youssef ; Scheichl, Robert. / A Stein variational Newton method. 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. editor / S Bengio ; H Wallach ; H Larochelle ; K Grauman ; N Cesa-Bianchi ; R Garnett. Vol. 2018-December San Diego CA USA, 2018. (Advances in Neural Information Processing Systems).

@inproceedings{16ea6bfa261b4676bbddfe4d54fff25a,

title = "A Stein variational Newton method",

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.",

author = "Gianluca Detommaso and Tiangang Cui and Alessio Spantini and Youssef Marzouk and Robert Scheichl",

year = "2018",

month = "1",

day = "1",

language = "English",

volume = "2018-December",

series = "Advances in Neural Information Processing Systems",

editor = "{ Bengio}, { S} and H Wallach and H Larochelle and K Grauman and Cesa-Bianchi, { N} and R Garnett",

booktitle = "Proceedings of Advances in Neural Information Processing Systems",

}

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, Advances in Neural Information Processing Systems, 2018, Montreal, Canada, 2/12/18.

A Stein variational Newton method. / Detommaso, Gianluca; Cui, Tiangang; Spantini, Alessio; Marzouk, Youssef; Scheichl, Robert.

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. ed. / S Bengio; H Wallach; H Larochelle; K Grauman; N Cesa-Bianchi; R Garnett. Vol. 2018-December San Diego CA USA, 2018. (Advances in Neural Information Processing Systems).

Research output: Chapter in Book/Report/Conference proceeding › Conference Paper › Research

TY - GEN

T1 - A Stein variational Newton method

AU - Detommaso, Gianluca

AU - Cui, Tiangang

AU - Spantini, Alessio

AU - Marzouk, Youssef

AU - Scheichl, Robert

PY - 2018/1/1

Y1 - 2018/1/1

N2 - 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.

AB - 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.

UR - http://www.scopus.com/inward/record.url?scp=85057843098&partnerID=8YFLogxK

M3 - Conference Paper

AN - SCOPUS:85057843098

VL - 2018-December

T3 - Advances in Neural Information Processing Systems

BT - Proceedings of Advances in Neural Information Processing Systems

A2 - Bengio, S

A2 - Wallach, H

A2 - Larochelle, H

A2 - Grauman, K

A2 - Cesa-Bianchi, N

A2 - Garnett, R

CY - San Diego CA USA

ER -

Detommaso G, Cui T, Spantini A, Marzouk Y, Scheichl R. A Stein variational Newton method. In Bengio S, Wallach H, Larochelle H, Grauman K, Cesa-Bianchi N, Garnett R, editors, 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. San Diego CA USA. 2018. (Advances in Neural Information Processing Systems).

Access to Document

274607717-oaFinal published version, 460 KB