Adversarial DEfense

Adversarial defense in machine learning aims to create models robust against adversarial attacks—maliciously crafted inputs designed to cause misclassification. Current research focuses on developing both training-based defenses, such as adversarial training and techniques leveraging optimal transport or energy-based models, and test-time defenses, including input preprocessing and model reprogramming. These efforts are crucial for ensuring the reliability and security of machine learning systems across diverse applications, from image classification and natural language processing to structural health monitoring and malware detection, where vulnerabilities could have significant consequences.