On-Average KL-Privacy and its equivalence to Generalization for Max-Entropy Mechanisms

Abstract

We define On-Average KL-Privacy and present its properties and connections to differential privacy, generalization and information-theoretic quantities including max-information and mutual information. The new definition significantly weakens differential privacy, while preserving its minimalistic design features such as composition over small group and multiple queries as well as closeness to post-processing. Moreover, we show that On-Average KL-Privacy is **equivalent** to generalization for a large class of commonly-used tools in statistics and machine learning that samples from Gibbs distributions---a class of distributions that arises naturally from the maximum entropy principle. In addition, a byproduct of our analysis yields a lower bound for generalization error in terms of mutual information which reveals an interesting interplay with known upper bounds that use the same quantity.

References in corpus (5)

• Rényi Divergence and Kullback-Leibler Divergence
• Scalable Privacy-Preserving Data Sharing Methodology for Genome-Wide Association Studies
• Privacy for Free: Posterior Sampling and Stochastic Gradient Monte Carlo
• Privacy and Statistical Risk: Formalisms and Minimax Bounds
• Preventing False Discovery in Interactive Data Analysis is Hard

Cited by in corpus (11)

• Information-theoretic analysis of generalization capability of learning algorithms
• Generating Artificial Data for Private Deep Learning
• Privacy-Preserving Dynamic Personalized Pricing with Demand Learning
• Three Tools for Practical Differential Privacy
• Privacy Risk in Machine Learning: Analyzing the Connection to Overfitting
• Déjà Vu: an empirical evaluation of the memorization properties of ConvNets
• Calibrating Noise to Variance in Adaptive Data Analysis
• Information, Privacy and Stability in Adaptive Data Analysis
• An Optimal Transport View on Generalization
• Per-instance Differential Privacy
• pMSE Mechanism: Differentially Private Synthetic Data with Maximal Distributional Similarity