Advances in Privacy-Preserving Machine Learning and Data Analysis

The field of privacy-preserving machine learning and data analysis is rapidly advancing, with a focus on developing innovative methods to protect sensitive information while maintaining the utility of the data. Recent research has led to the development of new techniques for differential privacy, including methods for optimizing canary sets to improve privacy auditing and frameworks for deriving generalization error bounds. Additionally, there have been significant advancements in local differential privacy, including the proposal of novel mechanisms that leverage correlations among attributes to improve utility while maintaining rigorous privacy guarantees. Other notable developments include the introduction of new algorithms for efficient neural network verification and the application of kernel sum of squares methods to global optimization problems. Noteworthy papers include: An Information-Theoretic Intersectional Data Valuation Theory, which introduces a formal pricing rule for quantifying and internalizing intersectional privacy loss, and Optimizing Canaries for Privacy Auditing with Metagradient Descent, which proposes a method for optimizing canary sets to improve privacy auditing. Furthermore, the paper Frequency Estimation of Correlated Multi-attribute Data under Local Differential Privacy presents a novel LDP mechanism that leverages correlations among attributes to substantially improve utility while maintaining rigorous LDP guarantees.

Sources

An Information-Theoretic Intersectional Data Valuation Theory

Sampling from Gaussian Processes: A Tutorial and Applications in Global Sensitivity Analysis and Optimization

A DPI-PAC-Bayesian Framework for Generalization Bounds

Optimizing Canaries for Privacy Auditing with Metagradient Descent

"We Need a Standard": Toward an Expert-Informed Privacy Label for Differential Privacy

Optimal Pure Differentially Private Sparse Histograms in Near-Linear Deterministic Time

A Privacy-Preserving Data Collection Method for Diversified Statistical Analysis

Efficient Neural Network Verification via Order Leading Exploration of Branch-and-Bound Trees

Frequency Estimation of Correlated Multi-attribute Data under Local Differential Privacy

KernelSOS for Global Sampling-Based Optimal Control and Estimation via Semidefinite Programming

Boosting Ray Search Procedure of Hard-label Attacks with Transfer-based Priors

Lower Bounds for Public-Private Learning under Distribution Shift

Minimax Data Sanitization with Distortion Constraint and Adversarial Inference

Built with on top of