Yes, but Did It Work?: Evaluating Variational Inference
Machine Learning
2018-10-15 v2 Computation
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.
Cite
@article{arxiv.1802.02538,
title = {Yes, but Did It Work?: Evaluating Variational Inference},
author = {Yuling Yao and Aki Vehtari and Daniel Simpson and Andrew Gelman},
journal= {arXiv preprint arXiv:1802.02538},
year = {2018}
}
Comments
Appearing at International Conference on Machine Learning 2018