Andrea Siposova

Hey there! I am Andrea, an AI researcher and data scientist with over 6 years of experience.

In the last 4 years, my research focus has been on the security and safety of AI and privacy-preserving machine learning. I spend my time developing and conducting experiments to enhance model security and data confidentiality. Having lived in 4 countries, I’ve learned to adapt quickly and approach problems from different angles. I’m motivated by creating solutions that genuinely help people, and I see advancing data security and privacy as meaningful work that I want to keep driving forward.

Along the way, I’ve been lucky enough to work with some great teams. Most recently, I worked within the Machine Learning and Data Management Group at SBA Research, where I focused on addressing challenges in the domain of secure and privacy-preserving machine learning. My work encompassed exploring data anonymization techniques, intentional memorization in ML models, as well as the mitigation of training data exfiltration. I have also developed defense applications implemented in a federated learning platform for healthcare applications FeatureCloud, which offer protection against exfiltration risks, while maintaining model utility.

Moreover, I was able to gain experience in computer vision and climate research. I led a climate research project gAia spanning 2.5 years, which was a part of the Austrian Security Research Programme. The project focused on landslide risk assessment from multimodal data, such as satellite images, climatology and weather data. Furthermore, as a researcher at WU (Vienna University of Economics and Business), I contributed to the development of advanced object detection and image segmentation approaches on satellite imagery within the ReKlaSAT3D project, a part of the Austrian Space Application Programme (ASAP).

As for my academic journey, I completed my Master’s in Data Science at TU Wien (Vienna University of Technology). In my diploma thesis I focused on the intersection of data privacy and the security of machine learning models. More specifically, I simulated and evaluated data exfiltration attacks on neural networks and developed defenses against them.

If you’re excited about the intersection of AI, security, and privacy as well, let’s chat! There’s still a lot to cover in this field, and I’m always up for a brainstorming session or collaboration.