Zero Knowledge in Machine Learning

Abstract: The usage of machine learning models is expanding at near-unprecedented rates, but the process by which models are produced is often opaque, with little known about the data on which they are trained, the reasons penalties and rewards are applied, or even the final number of parameters in the case of large commercial models. This talk will explore applications of zero knowledge in machine learning pipelines, with a particular focus on proving the quality of trained models.

Bio: Sarah Meiklejohn is a Professor of Cryptography and Security at University College London (UCL), affiliated with the Information Security Group in the Department of Computer Science, and a faculty member of the Initiative for Cryptocurrencies and Contracts (IC3), where she previously served as Associate Director. Her research focuses on computer security, cryptography, privacy-enhancing technologies, and cryptocurrencies. She previously worked as a Staff Research Scientist at Google and conducted research at Microsoft Research Redmond. She holds a Ph.D. in Computer Science from the University of California, San Diego, as well as an Sc.M. in Computer Science and an Sc.B. in Mathematics from Brown University.