Tools
New Concept of Defendability Against Backdoors in Machine Learning
Researchers from the Alignment Research Center have proposed a new formal concept of defendability against backdoors in machine learning models, framing it as a strategic game between an attacker and a defender.