arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 1603.04733stat.MLcs.LG
收藏

基于矩阵高斯后验的结构化高效变分深度学习

Structured and Efficient Variational Deep Learning with Matrix Gaussian Posteriors

University of Amsterdam(阿姆斯特丹大学);Canadian Institute for Advanced Research(加拿大高等研究院)

Christos Louizos, Max Welling

摘要:

我们介绍了一个变分贝叶斯神经网络的参数是通过随机矩阵的概率分布。具体来说,我们采用了一个矩阵变量高斯\cite{gupta1999matrix}参数后验分布,其中我们明确建模的协方差之间的输入和输出的每一层的尺寸。此外,近似的协方差阵,我们可以实现一个更有效的方式来表示这些相关性,也比完全分解的参数后验更便宜。我们进一步表明,与“本地reprarametrization技巧”\cite{kingma2015variational}在这个后验分布,我们到达一个高斯过程\cite{rasmussen2006gaussian}解释的隐藏单位在每一层,我们,同样与\cite{gal2015dropout},提供与深高斯过程的连接。我们继续利用这种二元性,并将“伪数据”\cite{snelson2005sparse}纳入我们的模型,这反过来又允许更有效的采样,同时保持原始模型的属性。通过大量的实验验证了该方法的有效性。

英文摘要:

We introduce a variational Bayesian neural network where the parameters are governed via a probability distribution on random matrices. Specifically, we employ a matrix variate Gaussian \cite{gupta1999matrix} parameter posterior distribution where we explicitly model the covariance among the input and output dimensions of each layer. Furthermore, with approximate covariance matrices we can achieve a more efficient way to represent those correlations that is also cheaper than fully factorized parameter posteriors. We further show that with the "local reprarametrization trick" \cite{kingma2015variational} on this posterior distribution we arrive at a Gaussian Process \cite{rasmussen2006gaussian} interpretation of the hidden units in each layer and we, similarly with \cite{gal2015dropout}, provide connections with deep Gaussian processes. We continue in taking advantage of this duality and incorporate "pseudo-data" \cite{snelson2005sparse} in our model, which in turn allows for more efficient sampling while maintaining the properties of the original model. The validity of the proposed approach is verified through extensive experiments.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑