Privacy-Aware Compression for Federated Learning

Explore privacy-aware compression techniques for federated learning in this Google TechTalk presented by Kamalika Chaudhuri at the 2022 Workshop on Federated Learning and Analytics. Delve into the challenges of federated learning, local differential privacy, and privacy mechanisms through a simple example. Examine the full algorithm, metric differential privacy, and interpolated MBU. Gain insights from Chaudhuri, a professor at UC San Diego and research scientist at Meta AI, as she shares her expertise in computer science and engineering. Conclude with a Q&A session to further understand the implications of privacy-aware compression in federated learning environments.

Introduction
Challenges in Federated Learning
Simple example
Local Differential Privacy
What do we need
Privacy Mechanism
Full Algorithm
Metric Differential Privacy
Interpolated MBU
Conclusion
Questions