Collaborative anomaly detection on blockchain from noisy sensor data

Abstract

This paper proposes a framework for collaborative anomaly detection on Blockchain. Taking condition-based management of industrial asset as a practical example, we extend the notion of Smart Contract, which has been implicitly assumed to be deterministic, to be able to handle noisy sensor data. By formalizing the task of collaborative anomaly detection as that of multi-task probabilistic dictionary learning, we show that major technical issues of validation, consensus building, and data privacy are naturally addressed within a statistical machine learning algorithm. We envision Blockchain as a platform for collaborative learning rather than just a traceable, immutable, and decentralized data management system, suggesting the direction towards 'Blockchain 3.0'.