Random Logit Scaling: Defending Deep Neural Networks Against Black-Box Score-Based Adversarial Example Attacks
Machine LearningArtificial IntelligenceCryptography and Security
Summary
The authors address the problem of tricky inputs, called adversarial examples, that can fool machine learning models, especially in settings where the attacker has limited information (black-box). They propose a new defense called Random Logit Scaling (RLS) that randomly alters model outputs to confuse attackers without harming accuracy. They show RLS reduces attack success more effectively and with less impact on confidence scores than other randomization methods. They also demonstrate that a popular defense, AAA, is vulnerable when attackers adapt their strategies accordingly.
Authors
Hamid Dashtbani, Mehdi Dousti Gandomani, AmirMahdi Sadeghzadeh
Abstract
Machine learning models are increasingly adapted in various domains. However, adversarial examples pose a significant threat to the reliable deployment of these models. In recent years, some powerful adversarial example attacks have been proposed for the fast and query-efficient generation of adversarial examples, even in black-box scenarios, highlighting the need for scalable, low-cost, and powerful defenses. In this work, we present two contributions to the domain of black-box adversarial example attacks and defenses. First, we propose Random Logit Scaling (RLS), a randomization-based defense against black-box score-based adversarial example attacks. RLS is a plug-and-play, post-processing defense that can be implemented on top of any existing ML model with minimal effort. The idea behind RLS is to confuse an attacker by outputting falsified scores resulting from randomly scaled logits while maintaining the model accuracy. We show that RLS significantly reduces the success rate of state-of-the-art black-box score-based attacks while preserving the accuracy and minimizing confidence score distortion compared to state-of-the-art randomization-based defenses. Second, we introduce a novel adaptive attack against AAA, a SOTA non-randomized black-box defense against black-box score-based attacks that also modifies output logits to confuse attackers, demonstrating its vulnerability against adaptive attacks.