Naoise Holohan - Diffprivlib: Privacy-preserving machine learning with Scikit-learn

Python

Explore the Diffprivlib library, a Python tool for implementing differential privacy in machine learning models using scikit-learn, with features like automatic parameter tuning and visualization.

Key takeaways

Differential privacy can be used to prevent attacks on anonymized datasets.
Diffprivlib is a Python library that provides tools for implementing differential privacy in machine learning models.
The library is designed to be easy to use and requires minimal expertise in differential privacy.
The library is built on top of NumPy and scikit-learn, making it easy to integrate with existing machine learning workflows.
Diffprivlib provides tools for several types of queries, including histograms, scalar queries, and vector queries.
The library also includes a mechanism for adding noise to the data to prevent re-identification.
The noise added is calibrated to a specific level of privacy, known as epsilon.
Epsilon is a measure of the amount of privacy that is sacrificed to achieve a specific level of accuracy.
The library provides a Python API for implementing differential privacy in machine learning models.
The models can be easily integrated with existing machine learning workflows using scikit-learn.
The library is designed to be flexible and can be used with a variety of machine learning algorithms.
The library is open-source and can be easily modified and extended to suit specific use cases.
Diffprivlib has several features that make it easy to use, including automatic parameter tuning and default parameterizations.
The library also provides tools for visualizing the results of differential privacy, such as histograms and scatter plots.
The library provides a way to track the privacy budget, which is the total amount of privacy that has been used.
The library provides a way to track the number of queries that have been executed.
The library provides a way to track the amount of data that has been accessed.
The library provides a way to track the number of requests that have been made.
Diffprivlib is a valuable tool for anyone who wants to implement differential privacy in their machine learning models.

Naoise Holohan - Diffprivlib: Privacy-preserving machine learning with Scikit-learn

More talks