BriefGPT.xyz
Nov, 2016
强健的变分推断
Robust Variational Inference
HTML
PDF
Michael Figurnov, Kirill Struminsky, Dmitry Vetrov
TL;DR
本文介绍了一种对变分推断的鲁棒修正方法，此方法适用于大多数训练集样本为随机噪声对象的情况，并在MNIST和OMNIGLOT合成数据集上展示了比证据下限更出色的目标。在原始的MNIST和OMNIGLOT数据集上，我们还观察到相对于非鲁棒证据的下限的小幅提高。
Abstract
variational inference
is a powerful tool for
approximate inference
. However, it mainly focuses on the
evidence lower bound
as variational
→