arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2305.10550cs.LGcond-mat.dis-nnq-bio.NC
收藏

无限宽深度神经网络中的稀疏性-深度权衡

Sparsity-depth Tradeoff in Infinitely Wide Deep Neural Networks

Weill Cornell Medical College of Cornell University(康奈尔大学威尔康奈尔医学院);Cornell Tech(康奈尔理工学院)

Chanwoo Chun, Daniel D. Lee

AI总结:

本文研究了宽度极限下稀疏神经活动对深度贝叶斯神经网络泛化性能的影响,推导出含ReLU和预定活跃神经元比例的NNGP核,发现浅层稀疏网络表现更优,并用核岭回归理论加以验证。

摘要:

我们研究了在宽度极限下,稀疏神经活动如何影响深度贝叶斯神经网络的泛化性能。为此,我们推导出了具有修正线性单元(ReLU)激活函数和预定比例活跃神经元的神经网络高斯过程(NNGP)核。使用NNGP核,我们观察到在多种数据集上,较浅深度的稀疏网络优于非稀疏网络。我们通过扩展核岭回归泛化误差的现有理论来验证这一观察。

英文摘要:

We investigate how sparse neural activity affects the generalization performance of a deep Bayesian neural network at the large width limit. To this end, we derive a neural network Gaussian Process (NNGP) kernel with rectified linear unit (ReLU) activation and a predetermined fraction of active neurons. Using the NNGP kernel, we observe that the sparser networks outperform the non-sparse networks at shallow depths on a variety of datasets. We validate this observation by extending the existing theory on the generalization error of kernel-ridge regression.

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。

↑