AAISM: Advanced in AI Security Management

Testing a visual-recognition model’s robustness against manipulated images directly assesses its susceptibility to adversarial examples and evasion attacks that can induce incorrect recognition results. NIST identifies evasion attacks and mitigations as central adversarial-machine-learning concerns.