Diffprivlib (the IBM Differential Privacy Library)

Diffprivlib (the IBM Differential Privacy Library) is an open-source, general-purpose Python library for differential privacy. Differential privacy protects individuals by adding carefully calibrated random noise to data analysis and machine learning, so that no single person's data can be identified from the results. The library lets researchers and developers experiment with differential privacy, explore its effect on machine learning and data analytics, and prototype their own privacy algorithms. Since its first release in 2019, it has been widely cited and used as a benchmark for new algorithms and libraries. The public release is intended for research and education only.

The library has four main components. Mechanisms are the building blocks of differential privacy, used by experts to build their own models. Models provide machine learning methods with built-in differential privacy, covering classification, clustering, regression, dimensionality reduction and data pre-processing. Tools provide differentially private versions of common data analysis functions, such as histograms. A budget accountant tracks the total privacy loss across multiple analyses.

Its models work like those of scikit-learn, a widely used machine learning library, so users can apply them with minimal changes to existing code. Users control the trade-off between privacy and accuracy with a parameter called epsilon. Example notebooks show, for instance, how model accuracy changes as privacy protection increases. The library is written in Python.

Auto-discovered on 2026-09-30 by OECD Catalogue Automation