BriefGPT.xyz
Jun, 2023
具结构离散表示的深度生成模型的无偏学习
Unbiased Learning of Deep Generative Models with Structured Discrete Representations
HTML
PDF
Harry Bendekgey, Gabriel Hope, Erik B. Sudderth
TL;DR
通过将图形模型与深度学习架构组合,我们学习具有两种框架优势的生成模型。我们提出了学习SVAE的新算法,并首次证明了SVAE处理丢失数据时处理多模态的能力。这些优化创新使SVAE能够通过梯度下降法进行学习。
Abstract
By composing
graphical models
with
deep learning
architectures, we learn generative models with the strengths of both frameworks. The
structured
→