Learning to Learn Task-Adaptive Hyperparameters for Few-Shot Learning
DC Field | Value | Language |
---|---|---|
dc.contributor.author | Baik, Sungyong | - |
dc.contributor.author | Choi, Myungsub | - |
dc.contributor.author | Choi, Janghoon | - |
dc.contributor.author | Kim, Heewon | - |
dc.contributor.author | Lee, Kyoung Mu | - |
dc.date.accessioned | 2024-04-09T02:00:23Z | - |
dc.date.available | 2024-04-09T02:00:23Z | - |
dc.date.issued | 2024-03 | - |
dc.identifier.issn | 0162-8828 | - |
dc.identifier.issn | 1939-3539 | - |
dc.identifier.uri | https://scholarworks.bwise.kr/ssu/handle/2018.sw.ssu/49454 | - |
dc.description.abstract | The objective of few-shot learning is to design a system that can adapt to a given task with only few examples while achieving generalization. Model-agnostic meta-learning (MAML), which has recently gained the popularity for its simplicity and flexibility, learns a good initialization for fast adaptation to a task under few-data regime. However, its performance has been relatively limited especially when novel tasks are different from tasks previously seen during training. In this work, instead of searching for a better initialization, we focus on designing a better fast adaptation process. Consequently, we propose a new task-adaptive weight update rule that greatly enhances the fast adaptation process. Specifically, we introduce a small meta-network that can generate per-step hyperparameters for each given task: learning rate and weight decay coefficients. The experimental results validate that learning a good weight update rule for fast adaptation is the equally important component that has drawn relatively less attention in the recent few-shot learning approaches. Surprisingly, fast adaptation from random initialization with ALFA can already outperform MAML. Furthermore, the proposed weight-update rule is shown to consistently improve the task-adaptation capability of MAML across diverse problem domains: few-shot classification, cross-domain few-shot classification, regression, visual tracking, and video frame interpolation. | - |
dc.format.extent | 14 | - |
dc.language | 영어 | - |
dc.language.iso | ENG | - |
dc.publisher | IEEE COMPUTER SOC | - |
dc.title | Learning to Learn Task-Adaptive Hyperparameters for Few-Shot Learning | - |
dc.type | Article | - |
dc.identifier.doi | 10.1109/TPAMI.2023.3261387 | - |
dc.identifier.bibliographicCitation | IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE, v.46, no.3, pp 1441 - 1454 | - |
dc.identifier.wosid | 001174191100023 | - |
dc.identifier.scopusid | 2-s2.0-85151570545 | - |
dc.citation.endPage | 1454 | - |
dc.citation.number | 3 | - |
dc.citation.startPage | 1441 | - |
dc.citation.title | IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE | - |
dc.citation.volume | 46 | - |
dc.identifier.url | https://ieeexplore.ieee.org/document/10080995 | - |
dc.publisher.location | 미국 | - |
dc.type.docType | Article | - |
dc.description.isOpenAccess | N | - |
dc.subject.keywordAuthor | Task analysis | - |
dc.subject.keywordAuthor | Optimization | - |
dc.subject.keywordAuthor | Mathematical models | - |
dc.subject.keywordAuthor | Adaptation models | - |
dc.subject.keywordAuthor | Visualization | - |
dc.subject.keywordAuthor | Training | - |
dc.subject.keywordAuthor | Neural networks | - |
dc.subject.keywordAuthor | Few-shot learning | - |
dc.subject.keywordAuthor | MAML | - |
dc.subject.keywordAuthor | meta-learning | - |
dc.subject.keywordAuthor | video frame interpolation | - |
dc.subject.keywordAuthor | visual tracking | - |
dc.relation.journalResearchArea | Computer Science | - |
dc.relation.journalResearchArea | Engineering | - |
dc.relation.journalWebOfScienceCategory | Computer Science, Artificial Intelligence | - |
dc.relation.journalWebOfScienceCategory | Engineering, Electrical & Electronic | - |
dc.description.journalRegisteredClass | scie | - |
dc.description.journalRegisteredClass | scopus | - |
Items in ScholarWorks are protected by copyright, with all rights reserved, unless otherwise indicated.
Soongsil University Library 369 Sangdo-Ro, Dongjak-Gu, Seoul, Korea (06978)02-820-0733
COPYRIGHT ⓒ SOONGSIL 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.