Current browse context:
stat.ML
Change to browse by:
References & Citations
Statistics > Machine Learning
Title: Variational Inference via $χ$-Upper Bound Minimization
(Submitted on 1 Nov 2016 (v1), last revised 12 Nov 2017 (this version, v4))
Abstract: Variational inference (VI) is widely used as an efficient alternative to Markov chain Monte Carlo. It posits a family of approximating distributions $q$ and finds the closest member to the exact posterior $p$. Closeness is usually measured via a divergence $D(q || p)$ from $q$ to $p$. While successful, this approach also has problems. Notably, it typically leads to underestimation of the posterior variance. In this paper we propose CHIVI, a black-box variational inference algorithm that minimizes $D_{\chi}(p || q)$, the $\chi$-divergence from $p$ to $q$. CHIVI minimizes an upper bound of the model evidence, which we term the $\chi$ upper bound (CUBO). Minimizing the CUBO leads to improved posterior uncertainty, and it can also be used with the classical VI lower bound (ELBO) to provide a sandwich estimate of the model evidence. We study CHIVI on three models: probit regression, Gaussian process classification, and a Cox process model of basketball plays. When compared to expectation propagation and classical VI, CHIVI produces better error rates and more accurate estimates of posterior variance.
Submission history
From: Adji Bousso Dieng [view email][v1] Tue, 1 Nov 2016 18:40:23 GMT (6045kb,D)
[v2] Mon, 27 Feb 2017 03:00:03 GMT (5663kb,D)
[v3] Mon, 6 Nov 2017 00:29:21 GMT (3841kb,D)
[v4] Sun, 12 Nov 2017 19:00:57 GMT (7729kb,D)
Link back to: arXiv, form interface, contact.