{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"indexed":{"date-parts":[[2026,4,14]],"date-time":"2026-04-14T00:22:03Z","timestamp":1776126123452,"version":"3.50.1"},"reference-count":39,"publisher":"Institute of Electrical and Electronics Engineers (IEEE)","issue":"4","license":[{"start":{"date-parts":[[2023,8,1]],"date-time":"2023-08-01T00:00:00Z","timestamp":1690848000000},"content-version":"vor","delay-in-days":0,"URL":"https:\/\/ieeexplore.ieee.org\/Xplorehelp\/downloads\/license-information\/IEEE.html"},{"start":{"date-parts":[[2023,8,1]],"date-time":"2023-08-01T00:00:00Z","timestamp":1690848000000},"content-version":"stm-asf","delay-in-days":0,"URL":"https:\/\/doi.org\/10.15223\/policy-029"},{"start":{"date-parts":[[2023,8,1]],"date-time":"2023-08-01T00:00:00Z","timestamp":1690848000000},"content-version":"stm-asf","delay-in-days":0,"URL":"https:\/\/doi.org\/10.15223\/policy-037"}],"funder":[{"name":"Huawei Technologies Co., Ltd","award":["HIRP2019041002010"],"award-info":[{"award-number":["HIRP2019041002010"]}]},{"name":"U.K. EPSRC","award":["EP\/P009727\/1"],"award-info":[{"award-number":["EP\/P009727\/1"]}]},{"DOI":"10.13039\/501100000275","name":"Leverhulme Trust","doi-asserted-by":"publisher","award":["RF-2019-492"],"award-info":[{"award-number":["RF-2019-492"]}],"id":[{"id":"10.13039\/501100000275","id-type":"DOI","asserted-by":"publisher"}]}],"content-domain":{"domain":[],"crossmark-restriction":false},"short-container-title":["IEEE Trans. Artif. Intell."],"published-print":{"date-parts":[[2023,8]]},"DOI":"10.1109\/tai.2022.3180272","type":"journal-article","created":{"date-parts":[[2022,6,9]],"date-time":"2022-06-09T16:41:17Z","timestamp":1654792877000},"page":"959-971","source":"Crossref","is-referenced-by-count":7,"title":["Delve Into Neural Activations: Toward Understanding Dying Neurons"],"prefix":"10.1109","volume":"4","author":[{"given":"Ziping","family":"Jiang","sequence":"first","affiliation":[{"name":"LIRA Center, Lancaster University, Lancaster, U.K."}]},{"given":"Yunpeng","family":"Wang","sequence":"additional","affiliation":[{"name":"Huawei Technologies Co., Ltd, Shenzhen, China"}]},{"ORCID":"https:\/\/orcid.org\/0000-0003-4735-6138","authenticated-orcid":false,"given":"Chang-Tsun","family":"Li","sequence":"additional","affiliation":[{"name":"Deakin University, Waurn Ponds, VIC, Australia"}]},{"ORCID":"https:\/\/orcid.org\/0000-0002-5770-934X","authenticated-orcid":false,"given":"Plamen","family":"Angelov","sequence":"additional","affiliation":[{"name":"LIRA Center, Lancaster University, Lancaster, U.K."}]},{"ORCID":"https:\/\/orcid.org\/0000-0003-1721-9474","authenticated-orcid":false,"given":"Richard","family":"Jiang","sequence":"additional","affiliation":[{"name":"LIRA Center, Lancaster University, Lancaster, U.K."}]}],"member":"263","reference":[{"key":"ref1","doi-asserted-by":"publisher","DOI":"10.1145\/3065386"},{"key":"ref2","doi-asserted-by":"publisher","DOI":"10.1016\/j.neunet.2014.09.003"},{"issue":"7553","key":"ref3","first-page":"436","article-title":"Deep learning","volume-title":"Nature","volume":"521","author":"LeCun","year":"2015"},{"key":"ref4","doi-asserted-by":"publisher","DOI":"10.1016\/0893-6080(89)90020-8"},{"key":"ref5","article-title":"Minimum width for universal approximation","volume-title":"Proc. Int. Conf. Learn. Representations","author":"Park","year":"2020"},{"key":"ref6","first-page":"249","article-title":"Understanding the difficulty of training deep feedforward neural networks","volume-title":"Proc. 13th Int. Conf. Artif. Intell. Statist.","volume":"9","author":"Glorot","year":"2010"},{"key":"ref7","first-page":"448","article-title":"Batch normalization: Accelerating deep network training by reducing internal covariate shift","volume-title":"Proc. Int. Conf. Mach. Learn.","author":"Ioffe","year":"2015"},{"key":"ref8","doi-asserted-by":"publisher","DOI":"10.1109\/ICCV.2015.123"},{"key":"ref9","article-title":"Sensitivity and generalization in neural networks: An empirical study","volume-title":"Proc. Int. Conf. Learn. Representations","author":"Novak","year":"2018"},{"key":"ref10","first-page":"315","article-title":"Deep sparse rectifier neural networks","volume-title":"Proc. 14th Int. Conf. Artif. Intell. Statist.","volume":"15","author":"Glorot","year":"2011"},{"key":"ref11","doi-asserted-by":"publisher","DOI":"10.5555\/3104322.3104425"},{"key":"ref12","article-title":"Empirical evaluation of rectified activations in convolutional network","author":"Xu","year":"2015"},{"key":"ref13","doi-asserted-by":"publisher","DOI":"10.4208\/cicp.OA-2020-0165"},{"key":"ref14","article-title":"Learning activation functions to improve deep neural networks","volume-title":"Proc. Int. Conf. Learn. Representations","author":"Agostinelli","year":"2015"},{"key":"ref15","article-title":"MobileNets: Efficient convolutional neural networks for mobile vision applications","author":"Howard","year":"2017"},{"key":"ref16","article-title":"BinaryConnect: Training deep neural networks with binary weights during propagations","volume-title":"Adv. Neural Inf. Process. Syst.","volume":"28","author":"Courbariaux","year":"2015"},{"key":"ref17","article-title":"Revise saturated activation functions","author":"Xu","year":"2016"},{"key":"ref18","article-title":"Gaussian error linear units (GELUs)","author":"Hendrycks","year":"2016"},{"key":"ref19","article-title":"Fast and accurate deep network learning by exponential linear units (ELUs)","volume-title":"Proc. Int. Conf. Learn. Representations (Poster)","author":"Clevert","year":"2016"},{"key":"ref20","article-title":"Searching for activation functions","author":"Ramachandran","year":"2017"},{"key":"ref21","first-page":"III-1310III-1318","article-title":"On the difficulty of training recurrent neural networks","volume-title":"Proc. 30th Int. Conf. Mach. Learn.","volume":"28","author":"Pascanu","year":"2013"},{"key":"ref22","article-title":"Gradient flow in recurrent nets: The difficulty of learning long-term dependencies","author":"Hochreiter","year":"2003","journal-title":"Field Guide Dyn. Recurrent Neural Netw."},{"key":"ref23","first-page":"21","article-title":"Layer normalization","volume-title":"Stat","volume":"1050","author":"Ba","year":"2016"},{"key":"ref24","first-page":"2488","article-title":"How does batch normalization help optimization?","volume-title":"Proc. 32nd Int. Conf. Neural Inf. Process. Syst.","author":"Santurkar","year":"2018"},{"key":"ref25","article-title":"A mean field theory of batch normalization","volume-title":"Proc. Int. Conf. Learn. Representations","author":"Yang","year":"2018"},{"key":"ref26","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.","volume":"89","author":"Kohler","year":"2019"},{"key":"ref27","doi-asserted-by":"publisher","DOI":"10.1007\/s10955-017-1836-5"},{"key":"ref28","first-page":"8572","article-title":"Wide neural networks of any depth evolve as linear models under gradient descent","volume-title":"Proc. 33rd Conf. Neural Inf. Process. Syst.","volume":"32","author":"Lee","year":"2019"},{"key":"ref29","doi-asserted-by":"publisher","DOI":"10.1109\/18.661502"},{"key":"ref30","first-page":"1517","article-title":"Benefits of depth in neural networks","volume-title":"Proc. 29th Annu. Conf. Learn. Theory","volume":"49","author":"Telgarsky","year":"2016"},{"key":"ref31","first-page":"6232","article-title":"The expressive power of neural networks: A view from the width","volume-title":"Proc. 31st Int. Conf. Neural Inf. Process. Syst.","author":"Lu","year":"2017"},{"key":"ref32","first-page":"2924","article-title":"On the number of response regions of deep feed forward networks with piece-wise linear activations","volume-title":"Adv. Neural Inf. Process. Syst.","author":"Pascanu","year":"2014"},{"key":"ref33","first-page":"2596","article-title":"Complexity of linear regions in deep networks","volume-title":"Proc. 36th Int. Conf. Mach. Learn.","volume":"97","author":"Hanin","year":"2019"},{"key":"ref34","first-page":"2924","article-title":"On the number of linear regions of deep neural networks","volume-title":"Proc. 27th Int. Conf. Neural Inf. Process. Syst.","volume":"2","author":"Montufar","year":"2014"},{"key":"ref35","first-page":"361","article-title":"Deep ReLU networks have surprisingly few activation patterns","volume-title":"Proc. 33rd Int. Conf. Neural Inf. Process. Syst.","author":"Hanin","year":"2019"},{"key":"ref36","article-title":"Spectrally-normalized margin bounds for neural networks","volume-title":"Proc. Int. Conf. Learn. Representations","author":"Bartlett","year":"2018"},{"key":"ref37","article-title":"A PAC-Bayesian approach to spectrally-normalized margin bounds for neural networks","volume-title":"Proc. Int. Conf. Learn. Representations","author":"Neyshabur","year":"2018"},{"key":"ref38","volume-title":"Geometric Measure Theory","author":"Federer","year":"2014"},{"key":"ref39","article-title":"Very deep convolutional networks for large-scale image recognition","author":"Simonyan","year":"2014"}],"container-title":["IEEE Transactions on Artificial Intelligence"],"original-title":[],"link":[{"URL":"http:\/\/xplorestaging.ieee.org\/ielx7\/9078688\/10190086\/09792293.pdf?arnumber=9792293","content-type":"unspecified","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2025,8,23]],"date-time":"2025-08-23T01:08:49Z","timestamp":1755911329000},"score":1,"resource":{"primary":{"URL":"https:\/\/ieeexplore.ieee.org\/document\/9792293\/"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2023,8]]},"references-count":39,"journal-issue":{"issue":"4"},"URL":"https:\/\/doi.org\/10.1109\/tai.2022.3180272","relation":{},"ISSN":["2691-4581"],"issn-type":[{"value":"2691-4581","type":"electronic"}],"subject":[],"published":{"date-parts":[[2023,8]]}}}