Few-Shot Learning on Graph Convolutional Network Based on Meta learning

被引：0

作者：

Liu X.-L. ^{[1
,4
]}

Feng L. ^{[1
]}

Liao L.-X. ^{[1
]}

Gong X. ^{[2
]}

Su H. ^{[1
]}

Wang J. ^{[3
]}

机构：

[1] College of Computer Science, Sichuan Normal University, Sichuan, Chengdu

[2] School of Computer and Artificial Intelligence, Southwest Jiao tong University, Sichuan, Chengdu

[3] School of Business, Sichuan Normal University, Sichuan, Chengdu

[4] Network Security Detachment, Ziyang Public Security Bureau, Sichuan, Ziyang

来源：

Tien Tzu Hsueh Pao/Acta Electronica Sinica | 2024年 / 52卷 / 03期

基金：

中国国家自然科学基金;

关键词：

convolutional neural network; few-shot learning; graph convolutional network; meta learning; prototype space;

D O I：

10.12263/DZXB.20220037

中图分类号：

学科分类号：

摘要：

Few shot learning is a hot and difficult problem in the field of machine learning. The existing few-shot learning model cannot effectively capture the relationships between data feature information and data label, thus causing the generalization ability of the resulting classifier would be weaker. A few-shot learning of graph convolutional network on prototype space, termed FSL-GCNPS, is developed. Firstly, the feature vectors are extracted on multi-task data by convolutional network. Secondly, in order to map the feature vectors into the prototype space, representation learning for the classes based on prototype network is proposed. Next, the graph is structured by combing the classes prototype vectors with class vectors. Then, FSL-GCNPS is trained using Meta learning. The experimental results show that FSL-GCNPS has better cross-domain adaptability in the medical image domain compared with the traditional deep learning models. Meanwhile, the FSL-GCNPS model has better classification accuracy and classification stability compared with the classical Few-shot learning algorithm. © 2024 Chinese Institute of Electronics. All rights reserved.

引用

页码：885 / 897

页数：12

共 32 条

[1] LECUN Y, BENGIO Y, HINTON G., Deep learning, Nature, 521, 7553, pp. 436-444, (2015)
[2] LI F, ZHAN B C, XIN L, Et al., Target recognition technology based on a new joint sensing matrix for compressed learning, Acta Electronica Sinica, 49, 11, pp. 2108-2116, (2021)
[3] SHI X J, CAO S, ZHANG C S, Et al., Research on automatic annotation algorithm for character-level oracle-bone images based on anchor points, Acta Electronica Sinica, 49, 10, pp. 2020-2031, (2021)
[4] XIE J L, ZHANG B T, LU Q., A dynamic head gesture recognition method based on 3D convolutional two-stream network fusion, Acta Electronica Sinica, 49, 7, pp. 1363-1369, (2021)
[5] LI F F, ROBERT F, PIETRO P., A Bayesian approach to unsupervised one-shot learning of object categories, Proceedings of the Ninth IEEE International Conference on Computer Vision, pp. 1134-1141, (2003)
[6] YU Y, FENG L, WANG G G, Et al., A few-shot learning model based on semi-supervised with pseudo label, Acta Electronica Sinica, 47, 11, pp. 2284-2291, (2019)
[7] DAI L C, FENG L, SHANG X L, Et al., Fast few-shot learning algorithm based on deep network, Pattern Recognition and Artificial Intelligence, 34, 10, pp. 941-956, (2021)
[8] VINYALS O, BLUNDELL C, LILLICRAP T, Et al., Matching networks for one shot learning, Proceedings of the 30th International Conference on Neural Information Processing Systems, pp. 3637-3645, (2016)
[9] SNELL J, SWERSKY K, ZEMEL R., Prototypical networks for few-shot learning, Proceedings of the 31st International Conference on Neural Information Processing Systems, pp. 4080-4090, (2017)
[10] SUNG F, YANG Y, ZHANG L, Et al., Learning to compare: Relation network for few-shot learning, Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pp. 1199-1208, (2018)

← 1 2 3 4 →