Machine learning algorithms are a fascinating set of tools that opened up completely new ways of analytics and evaluations. Modern neural networks outperform human capabilities when it comes to image recognition, reinforcement learning approaches allow computers to learn new tasks with incredible speed and many of the large scientific experiments would not be possible without the application of algorithmes that have learned to deal with the plethora of generated data.

I’m working at the Know-Center in Graz, a dedicated research center for machine learning applications and artificial intelligence. While I’m focussing especially on the combination of cryptography with machine learning to develop privacy-preserving analytics, the fields of research of my colleagues ranges from data management to natural language processing, from visual analytics to social computing and much more. Check out our website for further details.

There are also several privacy-preserving machine learning methods available that keep your data private even without cryptographic protocols. One example is federated learning, where a machine learning model (e.g. a neural network) is downloaded from a server or cloud and trained locally on several devices. After one forward pass the parameters of the model in the cloud get updated and the next local calculation starts. In the end you have a perfectly trained model with input from all devices without ever having to share any actual data, only the model parameter (e.g. the weights of the neural network) got transmitted and refined.