EIFFeL - Ensuring Integrity for Federated Learning

Explore the innovative EIFFeL system for ensuring both privacy and integrity in federated learning during this Google TechTalk presented by Amrita Roy Chowdhury from UC San Diego. Delve into the challenges of secure aggregation in federated learning and discover how EIFFeL addresses the issue of malformed updates designed to poison machine learning models. Learn about the framework's ability to enforce arbitrary integrity checks and remove malicious updates without compromising privacy. Gain insights into EIFFeL's practical applications, including its performance in training an MNIST classification model with 100 clients and 10% poisoning. Understand the importance of balancing update privacy and integrity in collaborative machine learning environments through this informative 31-minute presentation.