Differential Privacy as a Mutual Information Constraint

Abstract

Differential privacy is a precise mathematical constraint meant to ensure privacy of individual pieces of information in a database even while queries are being answered about the aggregate. Intuitively, one must come to terms with what differential privacy does and does not guarantee. For example, the definition prevents a strong adversary who knows all but one entry in the database from further inferring about the last one. This strong adversary assumption can be overlooked, resulting in misinterpretation of the privacy guarantee of differential privacy. Herein we give an equivalent definition of privacy using mutual information that makes plain some of the subtleties of differential privacy. The mutual-information differential privacy is in fact sandwiched between $\epsilon$-differential privacy and $(\epsilon ,\delta )$-differential privacy in terms of its strength. In contrast to previous works using unconditional mutual information, differential privacy is fundamentally related to conditional mutual information, accompanied by a maximization over the database distribution. The conceptual advantage of using mutual information, aside from yielding a simpler and more intuitive definition of differential privacy, is that its properties are well understood. Several properties of differential privacy are easily verified for the mutual information alternative, such as composition theorems.

ACM CCS 2016, 12 pages

References in corpus (3)

• Continuity of quantum conditional information
• Differential Privacy: on the trade-off between Utility and Information Leakage
• On the Continuity of the Secrecy Capacity of Compound and Arbitrarily Varying Wiretap Channels

Cited by in corpus (38)

• Deep Variational Information Bottleneck
• Coded Computing for Low-Latency Federated Learning over Wireless Edge Networks
• Privacy-Preserving Distributed Optimization via Subspace Perturbation: A General Framework
• An operational information decomposition via synergistic disclosure
• Robust Privacy-Utility Tradeoffs under Differential Privacy and Hamming Distortion
• Modelling and Quantifying Membership Information Leakage in Machine Learning
• Recent advances in deep learning theory
• Quantum Pufferfish Privacy: A Flexible Privacy Framework for Quantum Systems
• Adaptive Differentially Quantized Subspace Perturbation (ADQSP): A Unified Framework for Privacy-Preserving Distributed Average Consensus
• Utility-Optimized Local Differential Privacy Mechanisms for Distribution Estimation
• Privacy-Utility Trade-off of Linear Regression under Random Projections and Additive Noise
• Information-theoretic metrics for Local Differential Privacy protocols
• Toward Evaluating Re-identification Risks in the Local Privacy Model
• Pufferfish Privacy: An Information-Theoretic Study
• PRIVIC: A privacy-preserving method for incremental collection of location data
• Improved Information Theoretic Generalization Bounds for Distributed and Federated Learning
• Local Distribution Obfuscation via Probability Coupling
• Contraction of Private Quantum Channels and Private Quantum Hypothesis Testing
• Structure and Sensitivity in Differential Privacy: Comparing K-Norm Mechanisms
• Tighter Generalization Bounds for Iterative Differentially Private Learning Algorithms
• Obfuscation via Information Density Estimation
• Data Sanitisation Protocols for the Privacy Funnel with Differential Privacy Guarantees
• Multi-Server Weakly-Private Information Retrieval
• The Privacy Funnel from the viewpoint of Local Differential Privacy
• Robustness, Privacy, and Generalization of Adversarial Training
• Local Information Privacy and Its Application to Privacy-Preserving Data Aggregation
• Statistical Inference in the Differential Privacy Model
• The Cost of a Reductions Approach to Private Fair Optimization
• Compressive Privacy for a Linear Dynamical System
• An Extension of Fano's Inequality for Characterizing Model Susceptibility to Membership Inference Attacks
• Privacy-Preserving Distributed Processing: Metrics, Bounds, and Algorithms
• An Information-Theoretic Analysis of The Cost of Decentralization for Learning and Inference Under Privacy Constraints
• SoK on Performance Bounds in Anonymous Communication
• Simulation of Random Variables under Rényi Divergence Measures of All Orders
• Strong Privacy and Utility Guarantee: Over-the-Air Statistical Estimation
• Achieving Dalenius' Goal of Data Privacy with Practical Assumptions
• Low Influence, Utility, and Independence in Differential Privacy: A Curious Case of $\left(\genfrac{}{}{0ex}{}{3}{2}\right)$
• Improving Frequency Estimation under Local Differential Privacy