CS Special Seminar: Sebastian Szyller "Trustworthy Machine Learning: Niche to Mainstream"

Trustworthy Machine Learning: Niche to Mainstream

Sebastian Szyller
Intel Labs
Google Scholar

Abstract: In the last twenty years, machine learning (ML) based systems have transitioned from domain-specific expert tools to the front pages of all media. From early logistic regression to contemporary vision-language models, the democratisation of ML has reshaped the landscape of threats to models. While early ML systems dealt with primarily theoretical vulnerabilities, modern models face sophisticated attacks -- such as adversarial examples against computer vision, prompt injection, or model extraction -- that can result in real-world harm, and financial losses. Throughout this talk, we will explore the development of trustworthy ML from the complementary perspectives of the practitioner and the adversary. I will reveal how this dual viewpoint informed my past research, and the implications for the future of the field. 

Bio: Sebastian is a research scientist at Intel Labs. His research focuses on trustworthy and adversarial machine learning. Recently, he has been working on robustness, integrity, and provenance in generative models. He obtained his PhD from Aalto University, supervised by N. Asokan. His dissertation on "Ownership and Confidentiality in Machine Learning" received the Aalto Distinguished Dissertation Award, and the Finnish AI Society Dissertation Award.