LDP$^3$: An Extensible and Multi-Threaded Toolkit for Local Differential Privacy Protocols and Post-Processing Methods

2025-07-08

AI Summary

Existing local differential privacy (LDP) protocols and post-processing methods lack a unified, scalable evaluation framework. Method: We introduce the first open-source, modular, multithreaded LDP benchmarking toolkit, enabling flexible integration of mainstream and emerging LDP protocols (e.g., RAPPOR, OLH, Subset Selection) and post-processing techniques (e.g., frequency calibration, matrix inversion). It supports automated utility optimization across privacy budgets (ε), dataset sizes, and distributions. Contributions/Results: Leveraging parallelization and multi-metric evaluation (MAE, KL divergence, Top-k accuracy), the toolkit accelerates large-scale experiments significantly—achieving an 8.2× speedup on 16 cores. Empirical results show that optimal protocol–post-processing combinations improve data utility by 23.6% on average. This work establishes a standardized evaluation infrastructure and practical optimization paradigm for LDP research.

Technology Category

• Machine Learning: Privacy
• Search and Optimization: Distributed Search
• Natural Language Processing: Safety and Robustness

Application Category

• Security and Privacy: Large-scale security measurements
• Responsible Web: Data and user privacy-enhancing technologies for the Web
• Search and Retrieval-Augmented AI: Web evaluation methodologies and metrics

Abstract

Local differential privacy (LDP) has become a prominent notion for privacy-preserving data collection. While numerous LDP protocols and post-processing (PP) methods have been developed, selecting an optimal combination under different privacy budgets and datasets remains a challenge. Moreover, the lack of a comprehensive and extensible LDP benchmarking toolkit raises difficulties in evaluating new protocols and PP methods. To address these concerns, this paper presents LDP$^3$ (pronounced LDP-Cube), an open-source, extensible, and multi-threaded toolkit for LDP researchers and practitioners. LDP$^3$ contains implementations of several LDP protocols, PP methods, and utility metrics in a modular and extensible design. Its modular design enables developers to conveniently integrate new protocols and PP methods. Furthermore, its multi-threaded nature enables significant reductions in execution times via parallelization. Experimental evaluations demonstrate that: (i) using LDP$^3$ to select a good protocol and post-processing method substantially improves utility compared to a bad or random choice, and (ii) the multi-threaded design of LDP$^3$ brings substantial benefits in terms of efficiency.

Problem

• Selecting optimal LDP protocols and post-processing methods for varying privacy budgets and datasets
• Lack of comprehensive and extensible toolkit for benchmarking LDP techniques
• Need for efficient evaluation and integration of new LDP protocols and methods

Innovation

• Extensible toolkit for LDP protocols
• Modular design for easy integration
• Multi-threaded execution for efficiency