Machine Learning for Secure Device Personalization Using Blockchain
Kushal Singla, Joy Bose, Sharvil Katariya · 2018
Recently, there is a growing trend towards machine learning models that run on client devices such as smartphones, with constraints such as model size and time. Often, the user data on client devices is considered private and so cannot be uploaded to an external server for processing and running the model. Blockchain provides a new way of sharing data that is secure and decentralized. In this paper, we provide a way to use blockchain to run a machine learning model in a decentralized way using various nodes to compute part of the learning task. We apply our system in a smart home IoT setting to generate user customizations based on user activity prediction for IoT devices. We use distributed association rule mining to generate the rules of user activity from the device logs. We describe the system architecture and simulate the system using Ethereum based tools.