Gradient masking

Gradient masking is a technique used in machine learning to obscure or alter gradient information, often employed in adversarial machine learning to prevent model vulnerabilities from being exploited. It affects the ability of attackers to generate effective adversarial examples by hiding or manipulating gradient signals.

Read More →