Swarm meta learning

Xiao Tian*, Yuzhang Jiang, Hua Tianfield

*Corresponding author for this work

Research output: Chapter in Book/Report/Conference proceedingChapter (peer-reviewed)peer-review

Abstract

Swarm learning is a kind of decentralized machine learning. In this paper, we propose a new framework of decentralized collaborative learning, called swarm meta learning, by combining swarm learning with meta learning, blockchain, and federated learning. Nodes in the network of swarm meta learning can choose to cooperate according to business scenario requirements, and each node does not need to upload its local learned model. Our proposed framework is able to avoid data transfer, reduce communication costs and protect data privacy. We apply our proposed swarm meta learning framework in two scenarios with limited datasets. The experimental results show that swarm meta learning enables more independence and trust among data parties in multi-site cooperation scenarios with limited datasets, and achieves high accuracy whilst protecting privacy.
Original languageEnglish
Title of host publicationFederated and Transfer Learning
EditorsRoozbeh Razavi-Far, Boyu Wang, Matthew E. Taylor, Qiang Yang
Place of PublicationCham
PublisherSpringer
Pages167–183
Number of pages17
ISBN (Electronic)9783031117480
ISBN (Print)9783031117473
DOIs
Publication statusPublished - 2022

Publication series

NameAdaptation, Learning, and Optimization
Volume27
ISSN (Print)1867-4534
ISSN (Electronic)1867-4542

Keywords

  • Blockchain
  • Decentralized Collaborative Learning
  • Federated Learning
  • Meta Learning
  • Swarm Meta Learning

ASJC Scopus subject areas

  • Computer Science(all)

Fingerprint

Dive into the research topics of 'Swarm meta learning'. Together they form a unique fingerprint.

Cite this