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