A Le Cam Type Bound for Adversarial Learning and Applications
This work addresses the robustness problem in machine learning for applications requiring defense against adversarial attacks, providing foundational theoretical insights.
The paper tackles the problem of determining fundamental limits for adversarial learning without assuming specific attack methods or learning processes, and it applies general bounds to non-trivial learning problems with examples of common attacks.
Robustness of machine learning methods is essential for modern practical applications. Given the arms race between attack and defense methods, one may be curious regarding the fundamental limits of any defense mechanism. In this work, we focus on the problem of learning from noise-injected data, where the existing literature falls short by either assuming a specific attack method or by over-specifying the learning problem. We shed light on the information-theoretic limits of adversarial learning without assuming a particular learning process or attacker. Finally, we apply our general bounds to a canonical set of non-trivial learning problems and provide examples of common types of attacks.