BriefGPT.xyz
Feb, 2018
更紧的变分界并不一定更好
Tighter Variational Bounds are Not Necessarily Better
HTML
PDF
Tom Rainforth, Adam R. Kosiorek, Tuan Anh Le, Chris J. Maddison, Maximilian Igl...
TL;DR
本文提供理论和实证证据表明,使用更严格的证据下界(ELBO)可能会降低梯度估计器的信噪比,从而对学习推理网络产生不利影响,并介绍了三种新算法:偏重要性加权自动编码器(PIWAE),乘法重要性加权自动编码器(MIWAE)和组合重要性加权自动编码器(CIWAE)。同时,我们的结果表明,PIWAE可以同时改善推理和生成网络的训练。
Abstract
We provide theoretical and empirical evidence that using tighter evidence lower bounds (ELBOs) can be detrimental to the process of learning an
inference network
by reducing the
signal-to-noise ratio
of the gradi
→