Tutorial: Differential Privacy and Learning: The Tools, The Results, and The Frontier

Опубликовано 18 августа 2016, 18:41

When is working with private data safe, and when is it risky? Are the risks inherent to the computation? Widespread availability of detailed personal data makes understanding privacy necessaryΓÇöan exciting yet daunting challenge. Differential privacy provides a framework for understanding the tradeoff between the loss of privacy for those whose data are input to a computation and the accuracy of that computationΓÇÖs output. This tutorial will not assume familiarity with differential privacy. We will cover the necessary definitions, help build intuition, and introduce the basic differential privacy toolkit. We will then highlight some connections to learning in the existing differential privacy literature, and challenges and open problems for differentially private learning tasks.