Privacy and Security

Massive data collection required for training deep neural networks presents serious privacy issues. Centralized collection of photos, speech, and video from millions of individuals is ripe with privacy risks. Ideally, the learning algorithms would protect the privacy of users’ training data, by guaranteeing that the output model generalizes away from the specifics of any individual user. However, established machine learning algorithms make no such guarantee, and private, sensitive training data can be recovered from models. A model may implicitly store some of its training data, and careful analysis of the model may therefore reveal sensitive information. We aim to better understand vulnerabilities of machine learning models and to provide proper defense methods against adversarial attacks.