arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2001.00921stat.MLcs.LG
收藏

具有瓶颈的宽神经网络是深高斯过程

Wide Neural Networks with Bottlenecks are Deep Gaussian Processes

University of Tennessee(田纳西大学);Oak Ridge National Lab(橡树岭国家实验室)

Devanshu Agrawal, Theodore Papamarkou, Jacob Hinkle

摘要:

最近有很多关于神经网络的“宽极限”的工作,其中贝叶斯神经网络(BNN)被证明收敛到高斯过程(GP),因为所有隐藏层都被发送到无限宽。然而,这些结果不适用于需要一个或多个隐藏层保持窄的架构。在本文中,我们考虑了BNN的宽限制,其中一些隐藏层,称为“瓶颈”,在有限的宽度举行。其结果是GP的组合,我们称之为“瓶颈神经网络高斯过程“(瓶颈NNGP)。虽然很直观,但证明的微妙之处在于表明,网络组合的广泛限制实际上是限制GP的组合。我们还从理论上分析了一个单瓶颈NNGP,发现瓶颈诱导的依赖性之间的输出的多输出网络,持续通过极端后瓶颈深度,并防止核心的网络失去歧视的权力在极端后瓶颈深度。

英文摘要:

There has recently been much work on the "wide limit" of neural networks, where Bayesian neural networks (BNNs) are shown to converge to a Gaussian process (GP) as all hidden layers are sent to infinite width. However, these results do not apply to architectures that require one or more of the hidden layers to remain narrow. In this paper, we consider the wide limit of BNNs where some hidden layers, called "bottlenecks", are held at finite width. The result is a composition of GPs that we term a "bottleneck neural network Gaussian process" (bottleneck NNGP). Although intuitive, the subtlety of the proof is in showing that the wide limit of a composition of networks is in fact the composition of the limiting GPs. We also analyze theoretically a single-bottleneck NNGP, finding that the bottleneck induces dependence between the outputs of a multi-output network that persists through extreme post-bottleneck depths, and prevents the kernel of the network from losing discriminative power at extreme post-bottleneck depths.

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。

↑