Efficient Privacy-Preserving Machine Learning for Blockchain Network

Abstract

A blockchain is a trustworthy and secure decentralised and distributed network that can be used in many places, like banking, finance, insurance, healthcare, and business. Recently, many communities in blockchain networks want to use machine learning models to get useful information from the large amounts of data that each participant owns but is spread out in different places. Distributed machine learning (DML) for blockchain networks has been studied as a way to run a learning model without putting all the data in one place. Even though there have been a number of ideas, privacy and security have not been dealt with well enough. As we will show later, the architecture has flaws and is not as efficient as it could be. In this paper, we propose a DML model for a permission block chain that protects privacy and solves privacy, security, and performance problems in a structured way. As core primitives, we come up with a stochastic gradient descent method with different levels of privacy and an error-based aggregation rule. Our model can handle any kind of differentially private learning algorithm that needs to define non-deterministic functions. The proposed error-based aggregation rule is a good way to stop attacks by a malicious node that tries to make DML models less accurate. In a differentially private scenario, the results of our experiments show that our proposed model is more resistant to attacks from the outside than other aggregation rules. Lastly, we show that our proposed model is very useful because it is easy to understand and takes little time to process.