Overview
Explore a groundbreaking approach to differentially private convex optimization in this IEEE Symposium presentation. Delve into the Approximate Minima Perturbation algorithm, which leverages off-the-shelf optimizers without requiring hyperparameter tuning, making it ideal for practical deployment. Examine the extensive empirical evaluation of state-of-the-art algorithms for differentially private convex optimization across various benchmark and real-world datasets. Gain insights into the open-source implementations of these algorithms and their performance on nine public datasets, including four high-dimensional ones. Learn how to build useful predictive models while guaranteeing the privacy of sensitive data through differential privacy techniques in convex optimization tasks.
Syllabus
Towards Practical Differentialy Private Convex Optimization
Taught by
IEEE Symposium on Security and Privacy