Abstract:
Over the past decade, the rapid rise of machine learning has transformed countless domains, bringing major advances in efficiency, accuracy, and overall system performance. From healthcare to finance to cybersecurity, ML has consistently outperformed traditional solutions, fueled by breakthroughs in model architectures, access to large-scale datasets, and improved training methodologies. However, this success has also brought new risks. A growing body of research has shown that machine learning models are vulnerable to attacks that compromise their security and privacy, raising serious concerns for their deployment in safety-critical applications such as autonomous driving or cyberdefense. Beyond the models themselves, the broader supply chains that deliver ML-based solutions have also emerged as a source of potential vulnerabilities, requiring careful examination and robust mitigation strategies. This talk will focus specifically on machine learning supply chains, their different forms, the threats they pose, and the opportunities they unlock, following a kind of Jekyll and Hyde duality, showing that when looking beyond what we can see, one thing can be used both for good and for bad.
Bio:
Dorjan is an Assistant professor at the Department of Computer Science of Sapienza University of Rome. He obtained his PhD from the same university in February 2022. His research interests lie in the intersection between machine learning and security, including deep learning uses in security problems, (distributed) privacy-preserving machine learning, cyber-intelligent agents, and the application and incorporation of deep learning in the cybersecurity domain. Through the years, he has carried out research on assessing the resilience and robustness of state-of-the-art deep learning applications to security tasks and designing novel defenses with security and privacy in mind. At the same time, he has worked on improving current security mechanisms relying on safe, robust, and secure deployments of machine learning-based solutions.