{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"indexed":{"date-parts":[[2026,6,3]],"date-time":"2026-06-03T17:28:44Z","timestamp":1780507724938,"version":"3.54.1"},"reference-count":60,"publisher":"Institute of Electrical and Electronics Engineers (IEEE)","license":[{"start":{"date-parts":[[2024,1,1]],"date-time":"2024-01-01T00:00:00Z","timestamp":1704067200000},"content-version":"vor","delay-in-days":0,"URL":"https:\/\/ieeexplore.ieee.org\/Xplorehelp\/downloads\/license-information\/IEEE.html"},{"start":{"date-parts":[[2024,1,1]],"date-time":"2024-01-01T00:00:00Z","timestamp":1704067200000},"content-version":"stm-asf","delay-in-days":0,"URL":"https:\/\/doi.org\/10.15223\/policy-029"},{"start":{"date-parts":[[2024,1,1]],"date-time":"2024-01-01T00:00:00Z","timestamp":1704067200000},"content-version":"stm-asf","delay-in-days":0,"URL":"https:\/\/doi.org\/10.15223\/policy-037"}],"funder":[{"DOI":"10.13039\/501100001809","name":"National Natural Science Foundation of China","doi-asserted-by":"publisher","award":["62125109"],"award-info":[{"award-number":["62125109"]}],"id":[{"id":"10.13039\/501100001809","id-type":"DOI","asserted-by":"publisher"}]},{"DOI":"10.13039\/501100001809","name":"National Natural Science Foundation of China","doi-asserted-by":"publisher","award":["61931023"],"award-info":[{"award-number":["61931023"]}],"id":[{"id":"10.13039\/501100001809","id-type":"DOI","asserted-by":"publisher"}]},{"DOI":"10.13039\/501100001809","name":"National Natural Science Foundation of China","doi-asserted-by":"publisher","award":["61932022"],"award-info":[{"award-number":["61932022"]}],"id":[{"id":"10.13039\/501100001809","id-type":"DOI","asserted-by":"publisher"}]},{"DOI":"10.13039\/501100001809","name":"National Natural Science Foundation of China","doi-asserted-by":"publisher","award":["62371288"],"award-info":[{"award-number":["62371288"]}],"id":[{"id":"10.13039\/501100001809","id-type":"DOI","asserted-by":"publisher"}]},{"DOI":"10.13039\/501100001809","name":"National Natural Science Foundation of China","doi-asserted-by":"publisher","award":["62320106003"],"award-info":[{"award-number":["62320106003"]}],"id":[{"id":"10.13039\/501100001809","id-type":"DOI","asserted-by":"publisher"}]},{"DOI":"10.13039\/501100001809","name":"National Natural Science Foundation of China","doi-asserted-by":"publisher","award":["62301299"],"award-info":[{"award-number":["62301299"]}],"id":[{"id":"10.13039\/501100001809","id-type":"DOI","asserted-by":"publisher"}]},{"DOI":"10.13039\/501100001809","name":"National Natural Science Foundation of China","doi-asserted-by":"publisher","award":["T2122024"],"award-info":[{"award-number":["T2122024"]}],"id":[{"id":"10.13039\/501100001809","id-type":"DOI","asserted-by":"publisher"}]},{"DOI":"10.13039\/501100001809","name":"National Natural Science Foundation of China","doi-asserted-by":"publisher","award":["62120106007"],"award-info":[{"award-number":["62120106007"]}],"id":[{"id":"10.13039\/501100001809","id-type":"DOI","asserted-by":"publisher"}]},{"DOI":"10.13039\/501100001809","name":"National Natural Science Foundation of China","doi-asserted-by":"publisher","award":["62250055"],"award-info":[{"award-number":["62250055"]}],"id":[{"id":"10.13039\/501100001809","id-type":"DOI","asserted-by":"publisher"}]}],"content-domain":{"domain":[],"crossmark-restriction":false},"short-container-title":["IEEE Trans. Signal Process."],"published-print":{"date-parts":[[2024]]},"DOI":"10.1109\/tsp.2024.3410291","type":"journal-article","created":{"date-parts":[[2024,6,6]],"date-time":"2024-06-06T17:44:12Z","timestamp":1717695852000},"page":"2827-2841","source":"Crossref","is-referenced-by-count":12,"title":["On Centralization and Unitization of Batch Normalization for Deep ReLU Neural Networks"],"prefix":"10.1109","volume":"72","author":[{"ORCID":"https:\/\/orcid.org\/0000-0002-1682-4480","authenticated-orcid":false,"given":"Wen","family":"Fei","sequence":"first","affiliation":[{"name":"Department of Electronic Engineering, Shanghai Jiao Tong University, Shanghai, China"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/orcid.org\/0000-0003-2522-5778","authenticated-orcid":false,"given":"Wenrui","family":"Dai","sequence":"additional","affiliation":[{"name":"Department of Computer Science and Engineering, Shanghai Jiao Tong University, Shanghai, China"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/orcid.org\/0000-0003-2888-594X","authenticated-orcid":false,"given":"Chenglin","family":"Li","sequence":"additional","affiliation":[{"name":"Department of Electronic Engineering, Shanghai Jiao Tong University, Shanghai, China"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/orcid.org\/0000-0002-9694-9880","authenticated-orcid":false,"given":"Junni","family":"Zou","sequence":"additional","affiliation":[{"name":"Department of Computer Science and Engineering, Shanghai Jiao Tong University, Shanghai, China"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/orcid.org\/0000-0003-4552-0029","authenticated-orcid":false,"given":"Hongkai","family":"Xiong","sequence":"additional","affiliation":[{"name":"Department of Electronic Engineering, Shanghai Jiao Tong University, Shanghai, China"}],"role":[{"vocabulary":"crossref","role":"author"}]}],"member":"263","reference":[{"key":"ref1","article-title":"Gaussian error linear units","author":"Hendrycks","year":"2016"},{"key":"ref2","article-title":"Rectifier nonlinearities improve neural network acoustic models","volume-title":"Proc. ICML Workshop Deep Learn. Audio, Speech, Lang. Process.","author":"Maas","year":"2013"},{"key":"ref3","doi-asserted-by":"publisher","DOI":"10.1109\/TSP.2020.3039360"},{"key":"ref4","doi-asserted-by":"publisher","DOI":"10.1109\/TSP.2019.2904921"},{"key":"ref5","doi-asserted-by":"publisher","DOI":"10.1109\/TSP.2021.3094911"},{"key":"ref6","doi-asserted-by":"publisher","DOI":"10.5555\/3045118.3045167"},{"key":"ref7","first-page":"2483","article-title":"How does batch normalization help optimization","volume-title":"Proc. 32nd Int. Conf. Adv. Neural Inf. Process. Syst.","volume":"31","author":"Santurkar","year":"2018"},{"key":"ref8","first-page":"18387","article-title":"Batch normalization provably avoids ranks collapse for randomly initialised deep networks","volume-title":"Proc. Adv. Neural Inf. Process. Syst.","volume":"33","author":"Daneshmand","year":"2020"},{"key":"ref9","first-page":"6406","article-title":"The normalization method for alleviating pathological sharpness in wide neural networks","volume-title":"Proc. Adv. Neural Inf. Process. Syst.","volume":"32","author":"Karakida","year":"2019"},{"key":"ref10","doi-asserted-by":"publisher","DOI":"10.1007\/978-3-030-58536-5_23"},{"key":"ref11","article-title":"Towards understanding regularization in batch normalization","volume-title":"Proc. 7th Int. Conf. Learn. Representation","author":"Luo","year":"2019"},{"key":"ref12","article-title":"Theoretical analysis of auto rate-tuning by batch normalization","volume-title":"Proc. 7th Int. Conf. Learn. Representation","author":"Arora","year":"2019"},{"key":"ref13","first-page":"14544","article-title":"Reconciling modern deep learning with traditional optimization analyses: The intrinsic learning rate","volume-title":"Proc. 34th Conf. Adv. Neural Inf. Process. Syst.","volume":"33","author":"Li","year":"2020"},{"key":"ref14","first-page":"9233","article-title":"Fast mixing of stochastic gradient descent with normalization and weight decay","volume-title":"Proc. Adv. Neural Inf. Process. Syst.","volume":"34","author":"Li","year":"2022"},{"key":"ref15","article-title":"Instance normalization: The missing ingredient for fast stylization","author":"Ulyanov","year":"2016"},{"key":"ref16","article-title":"Layer normalization","author":"Ba","year":"2016"},{"key":"ref17","doi-asserted-by":"publisher","DOI":"10.1007\/978-3-030-01261-8_1"},{"key":"ref18","first-page":"901","article-title":"Weight normalization: A simple reparameterization to accelerate training of deep neural networks","volume-title":"Proc. Adv. Neural Inf. Process. Syst.","volume":"29","author":"Salimans","year":"2016"},{"key":"ref19","doi-asserted-by":"publisher","DOI":"10.1109\/ICCV.2017.305"},{"key":"ref20","first-page":"13 434","article-title":"Is normalization indispensable for training deep neural network?","volume-title":"Proc. Int. Conf. Adv. Neural Inf. Process. Syst.","volume":"33","author":"Shao","year":"2020"},{"key":"ref21","article-title":"Characterizing signal propagation to close the performance gap in unnormalized ResNets","volume-title":"Proc. 8th Int. Conf. Learn. Representation","author":"Brock","year":"2020"},{"key":"ref22","doi-asserted-by":"publisher","DOI":"10.1109\/TSP.2016.2546221"},{"key":"ref23","first-page":"322","article-title":"Fine-grained analysis of optimization and generalization for overparameterized two-layer neural networks","volume-title":"Proc. 36th Int. Conf. Mach. Learn.","author":"Arora","year":"2019"},{"key":"ref24","first-page":"6158","article-title":"Learning and generalization in overparameterized neural networks, going beyond two layers","volume-title":"Proc. 33rd Int. Conf. Adv. Neural Inf. Process. Syst.","volume":"32","author":"Allen-Zhu","year":"2019"},{"key":"ref25","doi-asserted-by":"publisher","DOI":"10.1007\/s10994-019-05839-6"},{"key":"ref26","article-title":"Gradient descent provably optimizes over-parameterized neural networks","volume-title":"Proc. 7th Int. Conf. Learn. Representation","author":"Du","year":"2019"},{"key":"ref27","first-page":"242","article-title":"A convergence theory for deep learning via over-parameterization","volume-title":"Proc. 36th Int. Conf. Mach. Learn.","author":"Allen-Zhu","year":"2019"},{"key":"ref28","article-title":"Demystifying batch normalization in ReLU networks: Equivalent convex optimization models and implicit regularization","volume-title":"Proc. 10th Int. Conf. Learn. Rep.","author":"Ergen","year":"2022"},{"key":"ref29","first-page":"806","article-title":"Exponential convergence rates for batch normalization: The power of length-direction decoupling in non-convex optimization","volume-title":"Proc. 22nd Int. Conf. Artif. Intell. Statist.","author":"Kohler","year":"2019"},{"key":"ref30","first-page":"7694","article-title":"Understanding batch normalization","volume-title":"Proc. 32nd Conf. Adv. Neural Inf. Process. Syst.","volume":"31","author":"Bjorck","year":"2018"},{"key":"ref31","doi-asserted-by":"publisher","DOI":"10.1109\/TSP.2017.2708039"},{"key":"ref32","first-page":"34 689","article-title":"Understanding the generalization benefit of normalization layers: Sharpness reduction","volume-title":"Proc. 36th Int. Conf. Adv. Neural Inf. Process. Syst.","volume":"35","author":"Lyu","year":"2022"},{"key":"ref33","article-title":"Fixup initialization: Residual learning without normalization","volume-title":"Proc. 7th Int. Conf. Learn. Representation","author":"Zhang","year":"2019"},{"key":"ref34","first-page":"19 964","article-title":"Batch normalization biases residual blocks towards the identity function in deep networks","volume-title":"Proc. Adv. Neural Inf. Process. Syst.","volume":"33","author":"De","year":"2020"},{"key":"ref35","article-title":"MimicNorm: Weight mean and last BN layer mimic the dynamic of batch normalization","author":"Fei","year":"2020"},{"key":"ref36","first-page":"249","article-title":"Understanding the difficulty of training deep feedforward neural networks","volume-title":"Proc. 13th Int. Conf. Artif. Intell. Statist.","author":"Glorot","year":"2010"},{"key":"ref37","doi-asserted-by":"publisher","DOI":"10.1109\/ICCV.2015.123"},{"key":"ref38","volume-title":"Fundamentals of Digital Image Processing","author":"Jain","year":"1989"},{"key":"ref39","first-page":"3360","article-title":"Exponential expressivity in deep neural networks through transient chaos","volume-title":"Proc. Adv. Neural Inf. Process. Syst.","volume":"29","author":"Poole","year":"2016"},{"key":"ref40","article-title":"Deep information propagation","volume-title":"Proc. 5th Int. Conf. Learn. Representation","author":"Schoenholz","year":"2017"},{"key":"ref41","first-page":"342","article-title":"Kernel methods for deep learning","volume-title":"Proc. Adv. Neural Inf. Process. Syst.","volume":"22","author":"Cho","year":"2009"},{"key":"ref42","first-page":"257","article-title":"Freeze and chaos: NTK views on DNN normalization, checkerboard and boundary artifacts","volume-title":"Proc. Math. Sci. Mach. Learn.","author":"Jacot","year":"2022"},{"key":"ref43","doi-asserted-by":"publisher","DOI":"10.1007\/978-3-030-86380-7_19"},{"key":"ref44","first-page":"8141","article-title":"On exact computation with an infinitely wide neural net","volume-title":"Proc. Adv. Neural Inf. Process. Syst.","volume":"32","author":"Arora","year":"2019"},{"key":"ref45","first-page":"12893","article-title":"On the inductive bias of neural tangent kernels","volume-title":"Proc. Adv. Neural Inf. Process. Syst.","volume":"32","author":"Bietti","year":"2019"},{"key":"ref46","first-page":"10462","article-title":"Disentangling trainability and generalization in deep neural networks","volume-title":"Proc. 37th Int. Conf. Mach. Learn.","author":"Xiao","year":"2020"},{"key":"ref47","article-title":"The large learning rate phase of deep learning: The catapult mechanism","author":"Lewkowycz","year":"2020"},{"key":"ref48","first-page":"11674","article-title":"Towards explaining the regularization effect of initial large learning rate in training neural networks","volume-title":"Proc. Adv. Neural Inf. Process. Syst.","volume":"32","author":"Li","year":"2019"},{"key":"ref49","article-title":"The two regimes of deep network training","author":"Leclerc","year":"2020"},{"key":"ref50","article-title":"Very deep convolutional networks for large-scale image recognition","volume-title":"Proc. 3rd Int. Conf. Learn. Representation","author":"Simonyan","year":"2015"},{"key":"ref51","doi-asserted-by":"publisher","DOI":"10.1109\/CVPR.2016.90"},{"key":"ref52","doi-asserted-by":"publisher","DOI":"10.1109\/CVPR.2018.00745"},{"key":"ref53","doi-asserted-by":"publisher","DOI":"10.1109\/CVPR.2018.00474"},{"key":"ref54","doi-asserted-by":"publisher","DOI":"10.1007\/978-3-030-01264-9_8"},{"key":"ref55","doi-asserted-by":"publisher","DOI":"10.1109\/CVPR.2018.00716"},{"key":"ref56","doi-asserted-by":"publisher","DOI":"10.1109\/ICCV.2017.324"},{"key":"ref57","first-page":"91","article-title":"Faster R-CNN: Towards real-time object detection with region proposal networks","volume-title":"Proc. Adv. Neural Inf. Process. Syst.","volume":"28","author":"Ren","year":"2015"},{"key":"ref58","doi-asserted-by":"publisher","DOI":"10.1007\/978-3-030-58452-8_37"},{"key":"ref59","doi-asserted-by":"publisher","DOI":"10.1007\/978-3-319-10602-1_48"},{"key":"ref60","article-title":"MMDetection: Open MMLab detection toolbox and benchmark","author":"Chen","year":"2019"}],"container-title":["IEEE Transactions on Signal Processing"],"original-title":[],"link":[{"URL":"http:\/\/xplorestaging.ieee.org\/ielx8\/78\/10347386\/10551636.pdf?arnumber=10551636","content-type":"unspecified","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2024,9,6]],"date-time":"2024-09-06T07:06:09Z","timestamp":1725606369000},"score":1,"resource":{"primary":{"URL":"https:\/\/ieeexplore.ieee.org\/document\/10551636\/"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2024]]},"references-count":60,"URL":"https:\/\/doi.org\/10.1109\/tsp.2024.3410291","relation":{},"ISSN":["1053-587X","1941-0476"],"issn-type":[{"value":"1053-587X","type":"print"},{"value":"1941-0476","type":"electronic"}],"subject":[],"published":{"date-parts":[[2024]]}}}