BETA-FL: Blockchain-Event Triggered Asynchronous Federated Learning in Supply Chains

EasyChair Preprint 10058, version 1

Abstract

BETA-FL provides a distributed federated learning framework implemented for supply chains by tightly integrating private-permissioned blockchain for the trusted alliance among various actors involved in the supply chain. With a trusted ledger as the moderator in federated learning workflow, our approach ensures protection against malicious backdoor attacks on performance from both server and clients. Additionally, our asynchronous training regime allows scalability to a large number of federated clients with small and constant delay caused due to an event-triggering scheme. We showcase the milk powder classification task as a potential use-case in the food supply chain to avoid food wastage. Finally, we facilitate a dedicated channel for regulatory bodies in our blockchain environment for inspections and audits pertaining to the functioning of the supply chain.

Keyphrases: Asynchronous Training, Blockchain, Decentralized Machine Learning, Federated Learning (FL), Hyperledger Fabric (HLF), supply chains

@booklet{EasyChair:10058,
  author    = {Mayank Gulati and Narges Dadkhah and Benedikt Groß and Gerhard Wunder and Jovan Glavonjic and Aleksandar Pavlovic and Aleksandar Tomcic},
  title     = {BETA-FL: Blockchain-Event Triggered Asynchronous Federated Learning in Supply Chains},
  howpublished = {EasyChair Preprint 10058},
  year      = {EasyChair, 2023}}