BNET : Batch Normalization With Enhanced Linear Transformation

Batch normalization (BN) is a fundamental unit in modern deep neural networks. However, BN and its variants focus on normalization statistics but neglect the recovery step that uses linear transformation to improve the capacity of fitting complex data distributions. In this paper, we demonstrate tha...

Ausführliche Beschreibung

Bibliographische Detailangaben
Veröffentlicht in:IEEE transactions on pattern analysis and machine intelligence. - 1979. - 45(2023), 7 vom: 09. Juli, Seite 9225-9232
1. Verfasser: Xu, Yuhui (VerfasserIn)
Weitere Verfasser: Xie, Lingxi, Xie, Cihang, Dai, Wenrui, Mei, Jieru, Qiao, Siyuan, Shen, Wei, Xiong, Hongkai, Yuille, Alan
Format: Online-Aufsatz
Sprache:English
Veröffentlicht: 2023
Zugriff auf das übergeordnete Werk:IEEE transactions on pattern analysis and machine intelligence
Schlagworte:Journal Article
LEADER 01000naa a22002652 4500
001 NLM355232901
003 DE-627
005 20231226063921.0
007 cr uuu---uuuuu
008 231226s2023 xx |||||o 00| ||eng c
024 7 |a 10.1109/TPAMI.2023.3235369  |2 doi 
028 5 2 |a pubmed24n1184.xml 
035 |a (DE-627)NLM355232901 
035 |a (NLM)37018583 
040 |a DE-627  |b ger  |c DE-627  |e rakwb 
041 |a eng 
100 1 |a Xu, Yuhui  |e verfasserin  |4 aut 
245 1 0 |a BNET  |b Batch Normalization With Enhanced Linear Transformation 
264 1 |c 2023 
336 |a Text  |b txt  |2 rdacontent 
337 |a ƒaComputermedien  |b c  |2 rdamedia 
338 |a ƒa Online-Ressource  |b cr  |2 rdacarrier 
500 |a Date Completed 06.06.2023 
500 |a Date Revised 06.06.2023 
500 |a published: Print-Electronic 
500 |a Citation Status PubMed-not-MEDLINE 
520 |a Batch normalization (BN) is a fundamental unit in modern deep neural networks. However, BN and its variants focus on normalization statistics but neglect the recovery step that uses linear transformation to improve the capacity of fitting complex data distributions. In this paper, we demonstrate that the recovery step can be improved by aggregating the neighborhood of each neuron rather than just considering a single neuron. Specifically, we propose a simple yet effective method named batch normalization with enhanced linear transformation (BNET) to embed spatial contextual information and improve representation ability. BNET can be easily implemented using the depth-wise convolution and seamlessly transplanted into existing architectures with BN. To our best knowledge, BNET is the first attempt to enhance the recovery step for BN. Furthermore, BN is interpreted as a special case of BNET from both spatial and spectral views. Experimental results demonstrate that BNET achieves consistent performance gains based on various backbones in a wide range of visual tasks. Moreover, BNET can accelerate the convergence of network training and enhance spatial information by assigning important neurons with large weights accordingly 
650 4 |a Journal Article 
700 1 |a Xie, Lingxi  |e verfasserin  |4 aut 
700 1 |a Xie, Cihang  |e verfasserin  |4 aut 
700 1 |a Dai, Wenrui  |e verfasserin  |4 aut 
700 1 |a Mei, Jieru  |e verfasserin  |4 aut 
700 1 |a Qiao, Siyuan  |e verfasserin  |4 aut 
700 1 |a Shen, Wei  |e verfasserin  |4 aut 
700 1 |a Xiong, Hongkai  |e verfasserin  |4 aut 
700 1 |a Yuille, Alan  |e verfasserin  |4 aut 
773 0 8 |i Enthalten in  |t IEEE transactions on pattern analysis and machine intelligence  |d 1979  |g 45(2023), 7 vom: 09. Juli, Seite 9225-9232  |w (DE-627)NLM098212257  |x 1939-3539  |7 nnns 
773 1 8 |g volume:45  |g year:2023  |g number:7  |g day:09  |g month:07  |g pages:9225-9232 
856 4 0 |u http://dx.doi.org/10.1109/TPAMI.2023.3235369  |3 Volltext 
912 |a GBV_USEFLAG_A 
912 |a SYSFLAG_A 
912 |a GBV_NLM 
912 |a GBV_ILN_350 
951 |a AR 
952 |d 45  |j 2023  |e 7  |b 09  |c 07  |h 9225-9232