Anonymization as homeomorphic data space transformation for privacy-preserving deep learning

Abstract

Industry 4.0 is largely data-driven nowadays. Owners of the data, on the one hand, want to get added value from the data by using remote artificial intelligence tools as services, on the other hand, they concern on privacy of their data within external premises. Ideal solution for this challenge would be such anonymization of the data, which makes the data safe in remote servers and, at the same time, leaves the opportunity for the machine learning algorithms to capture useful patterns from the data. In this paper, we take the problem of supervised machine learning with deep feedforward neural nets and provide an anonymization algorithm (based on the homeomorphic data space transformation), which guarantees privacy of the data and allows neural networks to learn successfully. We made several experiments to show how much the performance of the trained neural nets will suffer from the deepening of the anonymization power.