arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 1907.06194cs.LGcs.CVeess.IV
收藏

一种理解深度网络的分治方法

A Divide-and-Conquer Approach towards Understanding Deep Networks

Friedrich-Alexander University Erlangen-Nürnberg(埃尔朗根-纽伦堡弗里德里希-亚历山大大学);Erlangen Graduate School in Advanced Optical Technologies(埃尔朗根先进光学技术研究生院)

Weilin Fu, Katharina Breininger, Roman Schaffert, Nishant Ravikumar, Andreas Maier

摘要:

深度神经网络在包括医学图像分割在内的各个领域都取得了巨大的成功。然而,它们长期以来一直被批评为黑箱,因为解释,理解和纠正架构是困难的,因为没有深层神经网络设计的通用理论。以前,精确学习被提出来融合深度架构和传统方法。以这种方式构建的深度网络受益于原始的已知算子,具有更少的参数,并提高了可解释性。然而,它们并不能在所有应用中产生最先进的性能。在本文中,我们建议使用已知的运营商来分析深度网络,通过采用分而治之的策略来替换网络组件,同时保持其性能。视网膜血管分割的任务是为此目的进行研究。我们从一个高性能的U-Net开始,并通过一步一步的转换表明,我们能够将网络划分为已知运营商的模块。结果表明,可训练引导滤波器和Frangi滤波器的可训练版本的组合在U-Net水平上产生性能(AUC 0.974 vs. 0.972),参数大幅减少(111,536 vs. 9,575)。此外,经过训练的层可以映射回其原始算法解释,并使用标准的信号处理工具进行分析。

英文摘要:

Deep neural networks have achieved tremendous success in various fields including medical image segmentation. However, they have long been criticized for being a black-box, in that interpretation, understanding and correcting architectures is difficult as there is no general theory for deep neural network design. Previously, precision learning was proposed to fuse deep architectures and traditional approaches. Deep networks constructed in this way benefit from the original known operator, have fewer parameters, and improved interpretability. However, they do not yield state-of-the-art performance in all applications. In this paper, we propose to analyze deep networks using known operators, by adopting a divide-and-conquer strategy to replace network components, whilst retaining its performance. The task of retinal vessel segmentation is investigated for this purpose. We start with a high-performance U-Net and show by step-by-step conversion that we are able to divide the network into modules of known operators. The results indicate that a combination of a trainable guided filter and a trainable version of the Frangi filter yields a performance at the level of U-Net (AUC 0.974 vs. 0.972) with a tremendous reduction in parameters (111,536 vs. 9,575). In addition, the trained layers can be mapped back into their original algorithmic interpretation and analyzed using standard tools of signal processing.

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。

↑