A Experimental Study to Invariance of Several Groups Action to the Input of Residual Networks

CORC > 沈阳自动化研究所 > 中国科学院沈阳自动化研究所 > 机器人学研究室

	A Experimental Study to Invariance of Several Groups Action to the Input of Residual Networks
	Yang XY(杨秀义)1,2 ; Han Z(韩志)2 ; Tang YD(唐延东)2 ; Han JD(韩建达)2
	2018
会议日期	August 11-13, 2018
会议地点	Wuyishan, China
关键词	Invariance Residual network Group action
页码	1263-1267
英文摘要	We present a fresh new hypothesis that CNN is insensitive to some variants from input training data example, these variants relate to original training inputs by group actions, and will verify this hypothesis experimentally. This hypothesis means CNNs can generalize well on these variants when training on randomly generated training data and illuminates the paradox Why CNNs fit real and noise data and fail drastically when making predictions for noise data. Our findings suggest the study about generalization theory of CNNs should consider into these invariance of group actions to the input training data samples.
产权排序	1
会议录	Proceeding of the 2018 IEEE International Conference on Information and Automation
会议录出版者	IEEE
会议录出版地	New York
语种	英语
ISBN号	978-1-5386-8069-8
内容类型	会议论文
源URL	[http://ir.sia.cn/handle/173321/23839]
专题	沈阳自动化研究所_机器人学研究室
通讯作者	Han JD(韩建达)
作者单位	1.University of Chinese Academy of Sciences, Beijing 100049, China 2.State Key Laboratory of Robotics, Shenyang Institute of Automation, Chinese Academy of Sciences, Shenyang 110016, China
推荐引用方式 GB/T 7714	Yang XY,Han Z,Tang YD,et al. A Experimental Study to Invariance of Several Groups Action to the Input of Residual Networks[C]. 见:. Wuyishan, China. August 11-13, 2018.