FedACS: an Efficient Federated Learning Method Among Multiple Medical Institutions with Adaptive Client Sampling

With the development of deep learning, the neural network model trained by massive data is widely used in various fields, which improves production and living efficiency. However, the construction of a large open-source dataset is difficult due to commercial or privacy reasons, which limits the perf...

Full description

Saved in:
Bibliographic Details
Published in2021 14th International Congress on Image and Signal Processing, BioMedical Engineering and Informatics (CISP-BMEI) pp. 1 - 6
Main Authors Gu, Yunchao, Hu, Quanquan, Wang, Xinliang, Zhou, Zhong, Lu, Sixu
Format Conference Proceeding
LanguageEnglish
Published IEEE 23.10.2021
Subjects
Adaptation models
classification
Collaborative work
curriculum learning
Data privacy
Deep learning
federated learning
Hospitals
medical image analysis
Signal processing algorithms
Training
Online AccessGet full text

Cover

More Information
Summary:With the development of deep learning, the neural network model trained by massive data is widely used in various fields, which improves production and living efficiency. However, the construction of a large open-source dataset is difficult due to commercial or privacy reasons, which limits the performance of the deep learning model. In this paper, we focus on medical image analysis and explore the possibility of federated training for medical image classification in different hospitals. We propose an adaptive client sampling algorithm, which creatively applies the curriculum learning strategy to federated learning. Our proposed method can effectively reduce the communication overhead in federated learning, and provide technical support for deep learning training in cross-institutional and non-data sharing scenarios. Comparative experiments on CIFAR-10, CIFAR-100, and a chest X-Ray classification dataset show the effectiveness of the proposed algorithm.
DOI:10.1109/CISP-BMEI53629.2021.9624434