2018 (modified: 11 Nov 2022)ICML 2018Readers: Everyone
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...