DC Field | Value | Language |
---|---|---|
dc.contributor.author | Lu, Lu | ko |
dc.contributor.author | Shin, Yeonjong | ko |
dc.contributor.author | Su, Yanhui | ko |
dc.contributor.author | Karniadakis, George Em | ko |
dc.date.accessioned | 2022-07-06T02:00:31Z | - |
dc.date.available | 2022-07-06T02:00:31Z | - |
dc.date.created | 2022-07-06 | - |
dc.date.issued | 2020-11 | - |
dc.identifier.citation | COMMUNICATIONS IN COMPUTATIONAL PHYSICS, v.28, no.5, pp.1671 - 1706 | - |
dc.identifier.issn | 1815-2406 | - |
dc.identifier.uri | http://hdl.handle.net/10203/297250 | - |
dc.description.abstract | The dying ReLU refers to the problem when ReLU neurons become inactive and only output 0 for any input. There are many empirical and heuristic explanations of why ReLU neurons die. However, little is known about its theoretical analysis. In this paper, we rigorously prove that a deep ReLU network will eventually die in probability as the depth goes to infinite. Several methods have been proposed to alleviate the dying ReLU. Perhaps, one of the simplest treatments is to modify the initialization procedure. One common way of initializing weights and biases uses symmetric probability distributions, which suffers from the dying ReLU. We thus propose a new initialization procedure, namely, a randomized asymmetric initialization. We show that the new initialization can effectively prevent the dying ReLU. All parameters required for the new initialization are theoretically designed. Numerical examples are provided to demonstrate the effectiveness of the new initialization procedure. | - |
dc.language | English | - |
dc.publisher | GLOBAL SCIENCE PRESS | - |
dc.title | Dying ReLU and Initialization: Theory and Numerical Examples | - |
dc.type | Article | - |
dc.identifier.wosid | 000592624200003 | - |
dc.identifier.scopusid | 2-s2.0-85097467094 | - |
dc.type.rims | ART | - |
dc.citation.volume | 28 | - |
dc.citation.issue | 5 | - |
dc.citation.beginningpage | 1671 | - |
dc.citation.endingpage | 1706 | - |
dc.citation.publicationname | COMMUNICATIONS IN COMPUTATIONAL PHYSICS | - |
dc.identifier.doi | 10.4208/cicp.OA-2020-0165 | - |
dc.contributor.localauthor | Shin, Yeonjong | - |
dc.contributor.nonIdAuthor | Lu, Lu | - |
dc.contributor.nonIdAuthor | Su, Yanhui | - |
dc.contributor.nonIdAuthor | Karniadakis, George Em | - |
dc.description.isOpenAccess | N | - |
dc.type.journalArticle | Article | - |
dc.subject.keywordAuthor | Neural network | - |
dc.subject.keywordAuthor | Dying ReLU | - |
dc.subject.keywordAuthor | Vanishing/Exploding gradient | - |
dc.subject.keywordAuthor | Randomized asym-metric initialization | - |
dc.subject.keywordPlus | DEEP NEURAL-NETWORKS | - |
dc.subject.keywordPlus | ERROR | - |
Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.