Information-Geometric Inverse Distillation for Enhancing Adversarial Transferability
A new method improves the transferability of adversarial examples, potentially increasing the risk of black-box attacks on AI systems.
A new method improves the transferability of adversarial examples, potentially increasing the risk of black-box attacks on AI systems.
Auto-scored: The paper presents a technique that enhances adversarial transferability, a dual-use capability that could facilitate real-world attacks.