Send email Copy Email Address

Felix Koltermann

Our new Faculty: Machine Learning expert Dr. Franziska Boenisch

Dr. Franziska Boenisch's mission is to design machine learning in a way that benefits society while also protecting privacy. In doing so, the new CISPA Faculty is tackling one of the greatest technical challenges of our time. In an interview, Boenisch told us what led her to join CISPA, what her work focuses on and what challenges she sees.

It was an e-mail from CISPA Faculty Dr. Yang Zhang that drew Dr. Franziska Boenisch's attention to the final round of applications for CISPA's tenure-track program. Zhang is part of her network. "I had just started my postdoc in Canada, I wasn't searching yet and I actually wanted to stay there for at least another year," Boenisch says. "But in the end, the offer and the opportunity to work here were just too good. The bottom line of a postdoc is to get a professorship, and so it happened a year or two earlier than I had expected." That's how she came to start as Faculty at CISPA on September 1.

What she found interesting at CISPA was the interface between information security and machine learning. "I focus on the security and privacy of machine learning. That's why I think this place offers great collaborations and support in both directions to promote joint work." Her postdoc at the Vector Institute in Toronto, Canada, taught her to strive for excellence, which is also one of the unwavering principles of CISPA founding director Prof. Dr. Dr. h.c. Michael Backes. "In Toronto, I was also able to connect with many scientists working on the latest innovations in machine learning. And now that I have my own PhD students, I want to bring them into contact with each other, promote internships and also lectures here at the center," Boenisch explains her plans.

Research focus

What drives Boenisch is a great fascination with machine learning: "I think it's one of the most exciting technologies of our millennium because it helps us to solve so many problems with regard to science, society and the climate. But in many cases, machine learning methods cannot be used yet because, for example, in medicine or finance, there is always privacy protection to consider." It is Boenisch's mission to develop solutions for this. It is, however, a challenging task: "Guaranteeing privacy and making accurate predictions at the same time is incredibly difficult," she states.

One of Boenisch's research foci, derived from the handling of training data, concerns individualized privacy protection and its implementation in individualized ML models. "This is something that is often neglected. Many privacy applications tend to only define a protection level for an entire group, such as all patients in a hospital," she explains. But for her, the key point is that society is diverse and individual people have very different requirements when it comes to protecting their privacy. "This is what distinguishes my research in this area from that of many others. I want to focus on individuals and their different needs," she continues.


In order to close this research gap, Boenisch founded the SprintML Lab together with Dr. Adam Dziedzic, who also became CISPA Faculty on September 1. SprintML stands for "Secure, Private, Robust, Interpretable, and Trustworthy Machine Learning". Boenisch describes the aim of this research lab as follows: "Nowadays, many ML systems are primarily concerned with one question: How can we make the most accurate predictions? What we want is to continue to make the most accurate predictions while at the same time taking into account aspects that are important to society, such as privacy. " To achieve this, Boenisch and Dziedzic are looking for PhD students, postdocs as well as interns. Applicants should bring "ideally, a great enthusiasm for machine learning. But beyond that, just an interest in improving ML models for the benefit of society." Talking to Boenisch and witnessing her enthusiasm for the subject leaves no doubt that she will soon achieve her goal of a team of 10 researchers.