PRACTICAL IMPLEMENTATION OF BLOCKCHAIN ENABLED FEDERATED LEARNING
Date
2020-05
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
University of Delaware
Abstract
The Blockchain technology continues to attract much attention from an increas-
ing number of industries seeking to bene t from this revolutionary infrastructure. Some
of the strongest advantages of blockchain include temper-proof log of transactions and
security without a central authority. Both technologies of blockchain and federeted
learning have been studied and developed independently. Federated learning has been
pioneered by Google, which is known as vanilla federated learning. However, vanilla
federated learning is not without its issues. For one, it has the problem of a single point
of failure by relying on a central server. If the central server fails or get breached, the
entire system could collapse. Secondly, devices are not incentivized to share their data,
and di erent levels of qualities are not di erentiated at all, resulting in the reduction of
both quantity and quality of the device uploads. This objective of the thesis is to study
the feasibility of implementing Federated Learning with blockchain and to discover any
incompatibility issues to be resolved by practically implementing an example working
system in python. We demonstrated the functionality of blockchain-based federated
learning by running the python-based implementation of a linear regression model.
Experimental results validate the feasibility of such a system.
Description
Keywords
computer science, blockchain, federated learning