Explore a Google TechTalk presented by Marika Swanberg that delves into the privacy properties of attribution reporting methods through the lens of label inference measures. Gain insights into new adversary modeling techniques and two measures capturing different aspects of label inference success. Discover empirical and theoretical findings that guide discussions on risks and accuracy tradeoffs in potential attribution reporting methods. Learn about research conducted during Swanberg's Google PhD internship and Student Researcher position, in collaboration with Andres Muñoz Medina, Travis Dick, Robert Busa-Fekete, Claudio Gentile, and Adam Smith. Enhance your understanding of privacy concerns in data reporting and the complexities of balancing privacy protection with data utility in this 41-minute presentation from Boston University researcher Marika Swanberg.
A Unified Analysis of Label Inference Attacks
Overview
Syllabus
A Unified Analysis of Label Inference Attacks
Taught by
Google TechTalks
Related Courses
-
Machine Unlearning and Privacy Implications - Differential Privacy for ML
-
Low Cost High Power Membership Inference Attacks
-
Using Bayesian Inference to Detect Cyber Attacks
-
Differentially Private Sampling from Distributions - Google Algorithms Seminar
-
Membership Inference Attacks against Machine Learning Models
-
Are Your Sensitive Attributes Private? Novel Model Inversion Attribute Inference Attacks on Classification Models
