MVEB: Self-Supervised Learning With Multi-View Entropy Bottleneck
IEEE Transactions on Pattern Analysis and Machine Intelligence · 2024-09 · 46 (9) · 6097-6108
DOI: 10.1109/tpami.2024.3380065 ↗Kurz gefasst
Fachartikel in der Zeitschrift IEEE Transactions on Pattern Analysis and Machine Intelligence, 2024-09. Verfasst von Liangjian Wen und 3 weiteren Autor:innen. Die Arbeit wurde bisher 14-mal zitiert (Crossref) und nennt 58 Literaturangaben. Gefördert von: Zhejiang Lab, PCL, Shenzhen Science and Technology Innovation Commission.
Abstract (Originalsprache)
Self-supervised learning aims to learn representation that can be effectively generalized to downstream tasks. Many self-supervised approaches regard two views of an image as both the input and the self-supervised signals, assuming that either view contains the same task-relevant information and the shared information is (approximately) sufficient for predicting downstream tasks. Recent studies show that discarding superfluous information not shared between the views can improve generalization. Hence, the ideal representation is sufficient for downstream tasks and contains minimal superfluous information, termed minimal sufficient representation. One can learn this representation by maximizing the mutual information between the representation and the supervised view while eliminating superfluous information. Nevertheless, the computation of mutual information is notoriously intractable. In this work, we propose an objective termed multi-view entropy bottleneck (MVEB) to learn minimal sufficient representation effectively. MVEB simplifies the minimal sufficient learning to maximizing both the agreement between the embeddings of two views and the differential entropy of the embedding distribution. Our experiments confirm that MVEB significantly improves performance. For example, it achieves top-1 accuracy of 76.9% on ImageNet with a vanilla ResNet-50 backbone on linear evaluation. To the best of our knowledge, this is the new state-of-the-art result with ResNet-50.
Förderung
- Zhejiang Lab
- PCL
- Shenzhen Science and Technology Innovation Commission
Zitieren
APA
Wen, L., Wang, X., Liu, J., & Xu, Z. (2024). MVEB: Self-Supervised Learning With Multi-View Entropy Bottleneck. IEEE Transactions on Pattern Analysis and Machine Intelligence, 46(9), 6097-6108. https://doi.org/10.1109/tpami.2024.3380065
Vancouver
Wen L, Wang X, Liu J, Xu Z. MVEB: Self-Supervised Learning With Multi-View Entropy Bottleneck. IEEE Transactions on Pattern Analysis and Machine Intelligence. 2024;46(9):6097-6108. doi:10.1109/tpami.2024.3380065
BibTeX
@article{Wen2024,
author = {Wen, Liangjian and Wang, Xiasi and Liu, Jianzhuang and Xu, Zenglin},
title = {MVEB: Self-Supervised Learning With Multi-View Entropy Bottleneck},
journal = {IEEE Transactions on Pattern Analysis and Machine Intelligence},
year = {2024},
volume = {46},
number = {9},
pages = {6097--6108},
doi = {10.1109/tpami.2024.3380065}
}
EUVIMED