Machine Learning Model Hardening For Fun and Profit

Model Vulnerability

Machine learning models are vulnerable to adversarial attacks that cause misclassification. Hardening models improves robustness against attacks.

Adversarial Training

Training models on adversarial examples improves resistance to attacks. Exposing models to attacks during training builds robustness.

Evasion Resistance

Hardened models resist evasion attempts by adversaries. Techniques include adversarial training, input validation, and defensive distillation.

Practical Hardening

The presentation covers practical methods for hardening ML models in security applications. Model hardening makes systems more resistant to attacks and manipulation.