Explore a cutting-edge approach to measuring information leakage in black-box systems through this IEEE Symposium on Security & Privacy presentation. Dive into F-BLEAU (Fast Black-box Leakage Estimation), a novel method that leverages machine learning techniques to estimate Bayes risk and derive popular leakage measures. Learn how this approach overcomes limitations of traditional frequentist methods, particularly for systems with large or continuous output spaces. Discover the power of universally consistent learning rules, focusing on nearest neighbor rules, in improving estimation accuracy while reducing the number of required black-box queries. Examine the method's applicability through both synthetic and real-world data experiments, and compare its performance against the state-of-the-art tool leakiEst.
Overview
Syllabus
Introduction
Base risk
Nearest Neighbor
Results
Experiments
Taught by
IEEE Symposium on Security and Privacy
Tags
Related Courses
-
Targeted Adversarial Examples for Black Box Audio Systems
-
DP-Sniper - Black-Box Discovery of Differential Privacy Violations Using Classifiers
-
Measuring the Leakage of a Black-Box Using Machine Learning - Giovanni Cherubin
-
Black-Box Generation of Adversarial Text Sequences to Evade Deep Learning Classifiers
-
Hear "No Evil", See "Kenansville" - Efficient and Transferable Black-Box Attacks on Automatic Speech Recognition Systems
