Class Central is learner-supported. When you buy through links on our site, we may earn an affiliate commission.

YouTube

Gradient Obfuscation Gives a False Sense of Security in Federated Learning

USENIX via YouTube

Overview

Explore a critical analysis of privacy protection mechanisms in federated learning presented at USENIX Security '23. Delve into a new reconstruction attack framework for image classification tasks that challenges the effectiveness of gradient obfuscation techniques. Examine how common gradient postprocessing procedures, including quantization, sparsification, and perturbation, may provide a false sense of security. Discover a novel method for reconstructing images at the semantic level and learn about the quantification of semantic privacy leakage. Compare this approach with conventional image similarity scores and understand the implications for evaluating image data leakage in federated learning. Gain insights into the urgent need for revisiting and redesigning privacy protection mechanisms in existing federated learning algorithms to ensure robust client data security.

Syllabus

USENIX Security '23 - Gradient Obfuscation Gives a False Sense of Security in Federated Learning

Taught by

USENIX

Reviews

Start your review of Gradient Obfuscation Gives a False Sense of Security in Federated Learning

Never Stop Learning.

Get personalized course recommendations, track subjects and courses with reminders, and more.

Someone learning on their laptop while sitting on the floor.