Vector Institute Hosts Second Machine Learning Privacy and Security Workshop

The Vector Institute for Artificial Intelligence recently convened the second edition of its Machine Learning Privacy and Security Workshop. This event gathered faculty members, affiliates, postdoctoral fellows, and researchers to discuss innovations, emerging trends, and practical tools in ML security and privacy. The workshop aimed to foster a community focused on dependable and secure computing, addressing critical issues like robustness, privacy, and uncertainty quantification that hinder high-stakes real-world deployment. Key presentations included Pascale Gourdeau and Tosca Lechner discussing the computability of robust learning and adversarial attacks, highlighting how theoretical guarantees change when computational constraints are applied. Additionally, Reza Samavi presented on conformal prediction methods for deep neural networks, emphasizing the importance of uncertainty quantification in safety-critical applications such as medical diagnosis. The event underscored the need for collaborative efforts to solve recurring technical vulnerabilities and ensure technology advances responsibly, safely, and with preserved privacy.