On October 27, Alexandra Dmitrienko, a professor and Head of Secure Software Systems Research Group, Chair of Software Engineering (Informatik II), Department of Computer Science, at the University of Würzburg gave an exciting lecture about Safeguarding Security in Federated Machine Learning. The lecture was organized as part of the Female Scientists Lecture Series and was held at TU Darmstadt.
In her lecture, she conducted a comprehensive evaluation of the security and privacy risks associated with FL, delving into the specifics of targeted and untargeted poisoning attacks, as well as the countermeasures employed to mitigate these threats. In the discussion, Professor Dmitrienko highlighted the ongoing challenges in this field, such as the ability to differentiate between poisoned models and benign but uncommon models, particularly those trained on datasets with different data distributions. In addition to that, issues of adaptive attackers who, once aware of the detection method, can add an additional training loss to minimize any changes in the detection metric, effectively evading detection were discussed. Professor Dmitrienko’s lesson finished with potentials for improvements and outline promising directions to foster productive discussions in the research community.
Professor Dmitrienko also shared valuable career advice and personal insights about her career path in science.