Variational Auto-encoded Deep Gaussian Processes
Variational Auto-encoded Deep Gaussian Processes
复制标题
DOI：
--
复制
发表时间：
2015-11
期刊：
影响因子：
--
通讯作者：
Zhenwen Dai;Andreas C. Damianou;Javier I. González;Neil D. Lawrence
中科院分区：
文献类型：
--
作者：
Zhenwen Dai;Andreas C. Damianou;Javier I. González;Neil D. Lawrence
我们通过用识别模型增强深度高斯过程来开发一个可扩展的深度非参数生成模型。推理是在一个新的可扩展的变分框架中进行的，其中变分后验分布通过多层感知器重新参数化。这种重新表述的关键方面是，它可以防止扩散的变分参数，否则线性增长的样本大小成比例。我们推导出一种新的变分下界公式，使我们能够以一种能够处理主流深度学习任务大小的数据集的方式分配大部分计算。我们展示了该方法在各种挑战中的有效性，包括深度无监督学习和深度贝叶斯优化。
We develop a scalable deep non-parametric generative model by augmenting deep Gaussian processes with a recognition model. Inference is performed in a novel scalable variational framework where the variational posterior distributions are reparametrized through a multilayer perceptron. The key aspect of this reformulation is that it prevents the proliferation of variational parameters which otherwise grow linearly in proportion to the sample size. We derive a new formulation of the variational lower bound that allows us to distribute most of the computation in a way that enables to handle datasets of the size of mainstream deep learning tasks. We show the efficacy of the method on a variety of challenges including deep unsupervised learning and deep Bayesian optimization.