AAISM: Advanced in AI Security Management

Adversarial training improves a model’s resistance to adversarially manipulated inputs by training it on perturbed examples, reducing the likelihood that small, malicious input changes will cause evasion or misclassification.