Uploaded June 2025 | Updated September 2026, 3 weeks ago
By Nicolas Papernot, University of TorontoThe increasing integration of machine learning (ML) algorithms into critical aspects of our lives necessitates a fundamental understanding of what it means to "trust" these complex systems. This talk will delve into three key pillars essential for building and maintaining trust in ML. First, we will explore the crucial need for consistency with human expectations, examining the challenges posed by phenomena like adversarial examples. Second, we will address the necessity of auditability, focusing on an example drawn from the unlearning literature. Finally, we will discuss the critical importance of responsible and sustainable deployment, highlighting potential pitfalls such as model collapse and the broader ethical considerations for trustworthy ML.
By Nicolas Papernot, University of TorontoThe increasing integration of machine learning (ML) algorithms into critical aspects of our lives necessitates a fundamental understanding of what it means to "trust" these complex systems. This talk will delve into three key pillars essential for building and maintaining trust in ML. First, we will explore the crucial need for consistency with human expectations, examining the challenges posed by phenomena like adversarial examples. Second, we will address the necessity of auditability, focusing on an example drawn from the unlearning literature. Finally, we will discuss the critical importance of responsible and sustainable deployment, highlighting potential pitfalls such as model collapse and the broader ethical considerations for trustworthy ML.
