Deeply Explain CNN Via Hierarchical Decomposition

doi:10.1007/s11263-022-01746-x

	Deeply Explain CNN Via Hierarchical Decomposition
	Cheng, Ming-Ming 1; Jiang, Peng-Tao 1; Han, Ling-Hao 1; Wang, Liang2 ; Torr, Philip 3
发表期刊	INTERNATIONAL JOURNAL OF COMPUTER VISION
ISSN	0920-5691
	2023-01-11
页码	15
通讯作者	Cheng, Ming-Ming(cmm@nankai.edu.cn)
摘要	In computer vision, some attribution methods for explaining CNNs attempt to study how the intermediate features affect network prediction. However, they usually ignore the feature hierarchies among the intermediate features. This paper introduces a hierarchical decomposition framework to explain CNN's decision-making process in a top-down manner. Specifically, we propose a gradient-based activation propagation (gAP) module that can decompose any intermediate CNN decision to its lower layers and find the supporting features. Then we utilize the gAP module to iteratively decompose the network decision to the supporting evidence from different CNN layers. The proposed framework can generate a deep hierarchy of strongly associated supporting evidence for the network decision, which provides insight into the decision-making process. Moreover, gAP is effort-free for understanding CNN-based models without network architecture modification and extra training processes. Experiments show the effectiveness of the proposed method. The data and source code will be publicly available at https://mmcheng.net/hdecomp/.
关键词	Explaining CNNs Hierarchical decomposition
DOI	10.1007/s11263-022-01746-x
收录类别	SCI
语种	英语
资助项目	Major Project for New Generation of AI ; NSFC ; Fundamental Research Funds for the Central Universities (Nankai University) ; [2018AAA0100400] ; [61922046] ; [63223050]
项目资助者	Major Project for New Generation of AI ; NSFC ; Fundamental Research Funds for the Central Universities (Nankai University)
WOS研究方向	Computer Science
WOS类目	Computer Science, Artificial Intelligence
WOS记录号	WOS:000919320600002
出版者	SPRINGER
引用统计	被引频次：2[WOS] [WOS记录] [WOS相关记录]
文献类型	期刊论文
条目标识符	http://ir.ia.ac.cn/handle/173211/51351
专题	多模态人工智能系统全国重点实验室
通讯作者	Cheng, Ming-Ming
作者单位	1.Nankai Univ, TMCC, Tianjin, Peoples R China 2.NLPR, Beijing, Peoples R China 3.Univ Oxford, Oxford, England
推荐引用方式 GB/T 7714	Cheng, Ming-Ming,Jiang, Peng-Tao,Han, Ling-Hao,et al. Deeply Explain CNN Via Hierarchical Decomposition[J]. INTERNATIONAL JOURNAL OF COMPUTER VISION,2023:15.
APA	Cheng, Ming-Ming,Jiang, Peng-Tao,Han, Ling-Hao,Wang, Liang,&Torr, Philip.(2023).Deeply Explain CNN Via Hierarchical Decomposition.INTERNATIONAL JOURNAL OF COMPUTER VISION,15.
MLA	Cheng, Ming-Ming,et al."Deeply Explain CNN Via Hierarchical Decomposition".INTERNATIONAL JOURNAL OF COMPUTER VISION (2023):15.