Yes, but did it work?: Evaluating variational inference

Yuling Yao; Aki Vehtari; Daniel Simpson; Andrew Gelman

Yes, but did it work? Evaluating variational inference

Yuling Yao^*, Aki Vehtari, Daniel Simpson, Andrew Gelman

^*Corresponding author for this work

Research output: Chapter in Book/Report/Conference proceeding › Conference article in proceedings › Scientific › peer-review

28 Citations (Scopus)

107 Downloads (Pure)

Abstract

While it's always possible to compute a variational approximation to a posterior distribution, it can be difficult to discover problems with this approximation". We propose two diagnostic algorithms to alleviate this problem. The Paretosmoothed importance sampling (PSIS) diagnostic gives a goodness of fit measurement for joint distributions, while simultaneously improving the error in the estimate. The variational simulationbased calibration (VSBC) assesses the average performance of point estimates.

Original language	English
Title of host publication	35th International Conference on Machine Learning, ICML 2018
Editors	Jennifer Dy, Andreas Krause
Publisher	International Machine Learning Society
Pages	8887-8895
Number of pages	9
Volume	12
ISBN (Electronic)	9781510867963
Publication status	Published - 1 Jan 2018
MoE publication type	A4 Conference publication
Event	International Conference on Machine Learning - Stockholm, Sweden Duration: 10 Jul 2018 → 15 Jul 2018 Conference number: 35

Publication series

Name	Proceedings of Machine Learning Research
Publisher	PMLR
Number	80
ISSN (Electronic)	1938-7228

Conference

Conference	International Conference on Machine Learning
Abbreviated title	ICML
Country/Territory	Sweden
City	Stockholm
Period	10/07/2018 → 15/07/2018

Access to Document

yao18aFinal published version, 642 KB

http://proceedings.mlr.press/v80/yao18a.html

Cite this

@inproceedings{cda14f546e814974af57e1a87ed1f532,

title = "Yes, but did it work?: Evaluating variational inference",

abstract = "While it's always possible to compute a variational approximation to a posterior distribution, it can be difficult to discover problems with this approximation{"}. We propose two diagnostic algorithms to alleviate this problem. The Paretosmoothed importance sampling (PSIS) diagnostic gives a goodness of fit measurement for joint distributions, while simultaneously improving the error in the estimate. The variational simulationbased calibration (VSBC) assesses the average performance of point estimates.",

author = "Yuling Yao and Aki Vehtari and Daniel Simpson and Andrew Gelman",

year = "2018",

month = jan,

day = "1",

language = "English",

volume = "12",

series = "Proceedings of Machine Learning Research",

publisher = "International Machine Learning Society ",

number = "80",

pages = "8887--8895",

editor = "Jennifer Dy and Andreas Krause",

booktitle = "35th International Conference on Machine Learning, ICML 2018",

address = "United States",

note = "International Conference on Machine Learning, ICML ; Conference date: 10-07-2018 Through 15-07-2018",

}

Yao, Y, Vehtari, A, Simpson, D & Gelman, A 2018, Yes, but did it work? Evaluating variational inference. in J Dy & A Krause (eds), 35th International Conference on Machine Learning, ICML 2018. vol. 12, Proceedings of Machine Learning Research, no. 80, International Machine Learning Society , pp. 8887-8895, International Conference on Machine Learning, Stockholm, Sweden, 10/07/2018. <http://proceedings.mlr.press/v80/yao18a.html>

Yes, but did it work? Evaluating variational inference. / Yao, Yuling; Vehtari, Aki; Simpson, Daniel et al.
35th International Conference on Machine Learning, ICML 2018. ed. / Jennifer Dy; Andreas Krause. Vol. 12 International Machine Learning Society , 2018. p. 8887-8895 (Proceedings of Machine Learning Research; No. 80).

Research output: Chapter in Book/Report/Conference proceeding › Conference article in proceedings › Scientific › peer-review

TY - GEN

T1 - Yes, but did it work?

T2 - International Conference on Machine Learning

AU - Yao, Yuling

AU - Vehtari, Aki

AU - Simpson, Daniel

AU - Gelman, Andrew

N1 - Conference code: 35

PY - 2018/1/1

Y1 - 2018/1/1

N2 - While it's always possible to compute a variational approximation to a posterior distribution, it can be difficult to discover problems with this approximation". We propose two diagnostic algorithms to alleviate this problem. The Paretosmoothed importance sampling (PSIS) diagnostic gives a goodness of fit measurement for joint distributions, while simultaneously improving the error in the estimate. The variational simulationbased calibration (VSBC) assesses the average performance of point estimates.

AB - While it's always possible to compute a variational approximation to a posterior distribution, it can be difficult to discover problems with this approximation". We propose two diagnostic algorithms to alleviate this problem. The Paretosmoothed importance sampling (PSIS) diagnostic gives a goodness of fit measurement for joint distributions, while simultaneously improving the error in the estimate. The variational simulationbased calibration (VSBC) assesses the average performance of point estimates.

UR - http://www.scopus.com/inward/record.url?scp=85057318872&partnerID=8YFLogxK

M3 - Conference article in proceedings

AN - SCOPUS:85057318872

VL - 12

T3 - Proceedings of Machine Learning Research

SP - 8887

EP - 8895

BT - 35th International Conference on Machine Learning, ICML 2018

A2 - Dy, Jennifer

A2 - Krause, Andreas

PB - International Machine Learning Society

Y2 - 10 July 2018 through 15 July 2018

ER -