Toward Evolutionary Multitask Convolutional Neural Architecture Search
IEEE Transactions on Evolutionary Computation · 2024-06 · 28 (3) · 682-695
DOI: 10.1109/tevc.2023.3348475 ↗Kurz gefasst
Fachartikel in der Zeitschrift IEEE Transactions on Evolutionary Computation, 2024-06. Verfasst von Xun Zhou und 5 weiteren Autor:innen. Die Arbeit wurde bisher 40-mal zitiert (Crossref) und nennt 76 Literaturangaben. Gefördert von: National Key Research and Development Program of China, National Natural Science Foundation of China, Research Grants Council of the Hong Kong SAR.
Abstract (Originalsprache)
Evolutionary neural architecture search (ENAS) methods have been successfully used to design convolutional neural network (CNN) architectures automatically. These methods have achieved excellent performance in creating a specific neural architecture for a single task but are less efficient for multiple tasks. Existing ENAS frameworks always repeatedly perform the search from scratch for each task, even though these tasks may be solved by similar CNN architectures. This work presents an evolutionary multitask convolutional neural architecture search (MTNAS) framework to enable efficient architecture searches in multitask scenarios by incorporating architectural similarities. The proposed MTNAS constructs architectures for different tasks simultaneously by implementing a knowledgesharing mechanism among multiple search processes. Specifically, promising architectures found in one search process can be transferred and reused to generate high-quality architectures for others. Furthermore, we devise an adaptive strategy to dynamically adjust the frequency of knowledge transfer, aiming to alleviate the potential effect of negative transfer. Extensive experiments demonstrate that MTNAS can outperform stateof- the-art neural architecture search (NAS) methods or achieve comparable performance in different tasks but with $2\times $ less search cost.
Förderung
- National Key Research and Development Program of China
- National Natural Science Foundation of China
- Research Grants Council of the Hong Kong SAR
- Hong Kong Polytechnic University
- Shenzhen Technology Plan
- Characteristic Innovation Project of Colleges and Universities in Guangdong Province
Zitieren
APA
Zhou, X., Wang, Z., Feng, L., Liu, S., Wong, K., & Tan, K. C. (2024). Toward Evolutionary Multitask Convolutional Neural Architecture Search. IEEE Transactions on Evolutionary Computation, 28(3), 682-695. https://doi.org/10.1109/tevc.2023.3348475
Vancouver
Zhou X, Wang Z, Feng L, Liu S, Wong K, Tan KC. Toward Evolutionary Multitask Convolutional Neural Architecture Search. IEEE Transactions on Evolutionary Computation. 2024;28(3):682-695. doi:10.1109/tevc.2023.3348475
BibTeX
@article{Zhou2024,
author = {Zhou, Xun and Wang, Zhenkun and Feng, Liang and Liu, Songbai and Wong, Ka-Chun and Tan, Kay Chen},
title = {Toward Evolutionary Multitask Convolutional Neural Architecture Search},
journal = {IEEE Transactions on Evolutionary Computation},
year = {2024},
volume = {28},
number = {3},
pages = {682--695},
doi = {10.1109/tevc.2023.3348475}
}
EUVIMED