Yes, but Did It Work?: Evaluating Variational Inference
Y. Yao, A. Vehtari, D. Simpson, and A. Gelman. (2018)cite arxiv:1802.02538Comment: Appearing at International Conference on Machine Learning 2018.
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 Pareto-smoothed importance sampling (PSIS) diagnostic gives a goodness of
fit measurement for joint distributions, while simultaneously improving the
error in the estimate. The variational simulation-based calibration (VSBC)
assesses the average performance of point estimates.
Description
[1802.02538] Yes, but Did It Work?: Evaluating Variational Inference
%0 Journal Article
%1 yao2018evaluating
%A Yao, Yuling
%A Vehtari, Aki
%A Simpson, Daniel
%A Gelman, Andrew
%D 2018
%K bayesian variational
%T Yes, but Did It Work?: Evaluating Variational Inference
%U http://arxiv.org/abs/1802.02538
%X 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 Pareto-smoothed importance sampling (PSIS) diagnostic gives a goodness of
fit measurement for joint distributions, while simultaneously improving the
error in the estimate. The variational simulation-based calibration (VSBC)
assesses the average performance of point estimates.
@article{yao2018evaluating,
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 Pareto-smoothed importance sampling (PSIS) diagnostic gives a goodness of
fit measurement for joint distributions, while simultaneously improving the
error in the estimate. The variational simulation-based calibration (VSBC)
assesses the average performance of point estimates.},
added-at = {2020-02-13T03:12:29.000+0100},
author = {Yao, Yuling and Vehtari, Aki and Simpson, Daniel and Gelman, Andrew},
biburl = {https://www.bibsonomy.org/bibtex/2af628fe384081565ccdeca275c658632/kirk86},
description = {[1802.02538] Yes, but Did It Work?: Evaluating Variational Inference},
interhash = {511e52d110ea79995e62d76e4895bf78},
intrahash = {af628fe384081565ccdeca275c658632},
keywords = {bayesian variational},
note = {cite arxiv:1802.02538Comment: Appearing at International Conference on Machine Learning 2018},
timestamp = {2020-02-13T03:12:29.000+0100},
title = {Yes, but Did It Work?: Evaluating Variational Inference},
url = {http://arxiv.org/abs/1802.02538},
year = 2018
}