Incremental Learning with Adaptive Model Search and a Nominal Loss Model
DC Field | Value | Language |
---|---|---|
dc.contributor.author | Ahn, C. | - |
dc.contributor.author | Kim, Eunwoo | - |
dc.contributor.author | Oh, S. | - |
dc.date.accessioned | 2023-03-08T08:49:37Z | - |
dc.date.available | 2023-03-08T08:49:37Z | - |
dc.date.issued | 2022-02 | - |
dc.identifier.issn | 2169-3536 | - |
dc.identifier.uri | https://scholarworks.bwise.kr/cau/handle/2019.sw.cau/61689 | - |
dc.description.abstract | In incremental learning, different tasks are learned sequentially without access to the previously trained dataset. Catastrophic forgetting is a significant bottleneck to incremental learning as the network performs poorly on previous tasks when it is trained on a new task. This paper provides an adaptive model search method that uses a different part of parameters of the backbone network depending on an input image to mitigate catastrophic forgetting. Our model search method can prevent forgetting by minimizing the update of critical parameters for the previous tasks while learning a new task. This model search involves a trainable model search network that selects the model structure for an input image to minimize loss functions for all the tasks. We also propose a method for approximating the loss function of previous tasks using only the network parameters. The critical parameters for the previous tasks can be found according to the influence of the parameters on the approximated loss function. The proposed approximation method has the potential to theoretically reach a parameter set with a lower loss value than the parameter set learned by the previous task. The proposed framework is the first method of model search that can consider the performance of both current and previous tasks in the incremental learning problem. Empirical studies show that the proposed method outperforms other competitors for both old and new tasks while requiring less computation. Author | - |
dc.format.extent | 11 | - |
dc.language | 영어 | - |
dc.language.iso | ENG | - |
dc.publisher | Institute of Electrical and Electronics Engineers Inc. | - |
dc.title | Incremental Learning with Adaptive Model Search and a Nominal Loss Model | - |
dc.type | Article | - |
dc.identifier.doi | 10.1109/ACCESS.2022.3149598 | - |
dc.identifier.bibliographicCitation | IEEE Access, v.10, pp 16052 - 16062 | - |
dc.description.isOpenAccess | Y | - |
dc.identifier.wosid | 000756539700001 | - |
dc.identifier.scopusid | 2-s2.0-85124746164 | - |
dc.citation.endPage | 16062 | - |
dc.citation.startPage | 16052 | - |
dc.citation.title | IEEE Access | - |
dc.citation.volume | 10 | - |
dc.type.docType | Article | - |
dc.publisher.location | 미국 | - |
dc.subject.keywordAuthor | Adaptation models | - |
dc.subject.keywordAuthor | Artificial neural network | - |
dc.subject.keywordAuthor | Computational modeling | - |
dc.subject.keywordAuthor | computer vision | - |
dc.subject.keywordAuthor | Convolution | - |
dc.subject.keywordAuthor | incremental learning | - |
dc.subject.keywordAuthor | Learning systems | - |
dc.subject.keywordAuthor | model selection | - |
dc.subject.keywordAuthor | object recognition | - |
dc.subject.keywordAuthor | Search methods | - |
dc.subject.keywordAuthor | Task analysis | - |
dc.subject.keywordAuthor | Training | - |
dc.relation.journalResearchArea | Computer Science | - |
dc.relation.journalResearchArea | Engineering | - |
dc.relation.journalResearchArea | Telecommunications | - |
dc.relation.journalWebOfScienceCategory | Computer Science, Information Systems | - |
dc.relation.journalWebOfScienceCategory | Engineering, Electrical & Electronic | - |
dc.relation.journalWebOfScienceCategory | Telecommunications | - |
dc.description.journalRegisteredClass | scie | - |
dc.description.journalRegisteredClass | scopus | - |
Items in ScholarWorks are protected by copyright, with all rights reserved, unless otherwise indicated.
84, Heukseok-ro, Dongjak-gu, Seoul, Republic of Korea (06974)02-820-6194
COPYRIGHT 2019 Chung-Ang University All Rights Reserved.
Certain data included herein are derived from the © Web of Science of Clarivate Analytics. All rights reserved.
You may not copy or re-distribute this material in whole or in part without the prior written consent of Clarivate Analytics.