Cybersecure machine learning on open infrastructure

Modern machine learning offers opportunities for advanced data analysis in many different applications in society and industry, enabling more efficient responses to crises, improving processes and product safety, or decreasing environmental impact. Implementing such solutions on open infrastructure eliminates the need for expensive, dedicated infrastructure.

However, these opportunities are also accompanied by fundamental cybersecurity threats: Transferring data to the open infrastructure means that the data is subject to a (foreign) third-party’s security measures and becomes vulnerable to data breaches, espionage, and possibly foreign legislation.This project addresses this challenge by developing secure, privacy-preserving machine learning methods that ensure full data protection, preventing espionage, and ensuring full privacy.

This is realized by exploiting homomorphic encryption schemes and differential privacy together with state-of-the-art machine learning methods.

Furthermore, we implement the algorithms in an open source computing platform and demonstrate their applicability in a decentralized, collaborative localization and mapping task.This enables secure deployment of state-of-the-art machine learning on open infrastructure without compromising privacy and integrity, ensuring the full potential of the latest machine learning methods can be leveraged.