Explore the Matrix-Variate Gaussian (MVG) mechanism, a novel approach to differential privacy for matrix-valued query functions. Learn about the limitations of extending traditional mechanisms like Laplace and Gaussian to matrix-valued queries, and discover how the MVG mechanism addresses these challenges. Delve into key concepts including differential privacy, simple addition, metrics, and various methods for achieving privacy. Examine the MVT distribution, SVD, and the main theorem behind the MVG mechanism. Analyze experimental results and canonical queries to understand the practical applications of this innovative approach. This 23-minute video presentation from the Association for Computing Machinery (ACM) offers a comprehensive overview of the MVG mechanism and its implications for enhancing privacy in matrix-valued data analysis.
MVG Mechanism - Differential Privacy under Matrix-Valued Query
Association for Computing Machinery (ACM) via YouTube
Overview
Syllabus
Introduction
Differential Privacy
Simple Addition
Why Metrics
How to do Differential Privacy
Different methods for Differential Privacy
Outline
Key Idea
MVT Distribution
MVG Mechanism
SVD
Main Theorem
Conceptual Plot
Experiments
canonical query
experimental results
conclusion
Discussion
Taught by
Association for Computing Machinery (ACM)
