Federated Learning using Distributed Messaging with Entitlements for
Anonymous Computation and Secure Delivery of Model
- Monik Raj Behera ,
- sudhir upadhyay ,
- Robert Otter ,
- Suresh Shetty
Abstract
Federated learning has become one of the most recent and widely
researched areas of machine learning. Several machine-learning
frameworks, such as Tensorflow Federated and PySyft and others have
gained momentum in recent past and continue to evolve. Some of the
frameworks involve techniques such as differential privacy, secure
multi-party computation, gradient descent calculation over the network
to achieve privacy of underlying data in federated learning. While these
frameworks serve the need for a general-purpose federated learning model
as per certain framework, in this paper we present a solution based on
distributed messaging with appropriate entitlements that enterprises can
leverage in a managed and permissioned network. The solution implements
access controls on message source and destination in a decentralized
network, which can implement any given data science model in the
federated network to facilitate secure federated learning.