GSM: Global Semantic Memory

doi:10.1016/j.patcog.2025.111950

	GSM: Global Semantic Memory
	Li, Jiahao 1,2,3,4; Chen, Yiqiang 1,2,3; Xing, Yunbing 1,3,4; Gu, Yang 1,2,3; Lan, Xiangyuan 2
	2026
发表期刊	PATTERN RECOGNITION
ISSN	0031-3203
卷号	169 页码:16
摘要	Unsupervised anomaly detection is to detect previously unseen rare samples without any prior knowledge about them. With the emergence of deep learning, many methods employ normal data reconstruction to train detection models, which is expected to yield relatively large errors when reconstructing anomalies. However, recent studies find that anomalies can be overgeneralized, resulting in reconstruction errors as small as normal samples. In this paper, we examine the anomaly overgeneralization problem and propose global semantic information learning. Normal and anomalous samples may share the same local feature such as textures, edges, and corners, but have separability at the global semantic level. To address this, we propose the global semantic memory with novel cascade architecture designed to capture global semantic information in the latent space and introduce a configurable sparsification and random forgetting mechanism. Our proposed method achieves state-of-the-art experimental results on different public benchmarks, without the introduction of any additional auxiliary loss.
关键词	Anomaly detection Unsupervised learning Memory
DOI	10.1016/j.patcog.2025.111950
收录类别	SCI
语种	英语
资助项目	Innovation Funding of ICT, CAS[E461020] ; Hunan Provincial Natural Science Foundation of China[2023JJ70009] ; Beijing Municipal Science & Technology Commission[Z221100002722009] ; Youth Innovation Promotion Association CAS[2021101] ; Science and Technology Innovation Program of Hunan Province[2022RC4006] ; Project of Pengcheng Laboratory[PCL2023A08]
WOS研究方向	Computer Science ; Engineering
WOS类目	Computer Science, Artificial Intelligence ; Engineering, Electrical & Electronic
WOS记录号	WOS:001532675700001
出版者	ELSEVIER SCI LTD
引用统计
文献类型	期刊论文
条目标识符	http://119.78.100.204/handle/2XEOYT63/42077
专题	中国科学院计算技术研究所期刊论文_英文
通讯作者	Chen, Yiqiang
作者单位	1.Chinese Acad Sci, Inst Comp Technol, Beijing Key Lab Mobile Comp & Pervas Device, Beijing 100190, Peoples R China 2.Pengcheng Lab, Shenzhen 518000, Peoples R China 3.Univ Chinese Acad Sci, Beijing 100190, Peoples R China 4.AIER Eye Hosp Grp Co Ltd, Changsha 410015, Peoples R China
推荐引用方式 GB/T 7714	Li, Jiahao,Chen, Yiqiang,Xing, Yunbing,et al. GSM: Global Semantic Memory[J]. PATTERN RECOGNITION,2026,169:16.
APA	Li, Jiahao,Chen, Yiqiang,Xing, Yunbing,Gu, Yang,&Lan, Xiangyuan.(2026).GSM: Global Semantic Memory.PATTERN RECOGNITION,169,16.
MLA	Li, Jiahao,et al."GSM: Global Semantic Memory".PATTERN RECOGNITION 169(2026):16.