Data Protection for AI

In current AI development, the method of centralized machine learning is very common, where a statistical model is trained centrally on a server. A disadvantage of this method from the viewpoint of cybersecurity is that the centralized server creates a central data collection and a single point of attack. There are however ways to combine data use and privacy protection. One approach is distributed machine learning. “Distributed Machine Learning opens up new possibilities for effective and scalable use of data without having to share it. This enables many useful applications with sensitive data in the first place”, says professor Ahmad-Reza Sadeghi, head of the System Security Lab.

In distributed machine learning, each end device accesses the current training model and trains it locally with its own data set. The new publication “AI at a Glance. Distributed Machine Learning” (opens in new tab) by Lernende Systeme, Germany's Platform for Artificial Intelligence, explains how different methods of distributed machine learning work and their pros and cons with regard to data protection. Sadeghi, member of their working group on IT Security and Privacy, was one of the expert advisors for the publication. You can also read the related interview “Does AI need a new cybersecurity mindset?” with professor Sadeghi by online journalist Matthias Bastian at THE DECODER.