@ericyocam6Shipped: Adversarial AI — Defensive Models and Robustness Experiments
An open-source collection of defensive AI research artifacts supporting Book 3 of The Applied AI Universe Coding Guide series.
The project examines how machine learning systems can be protected against adversarial attacks and how those defenses perform when attackers adapt.
The associated research framework covers 35 classical and hybrid/quantum victims, spanning machine learning, deep neural networks, reinforcement learning, generative AI, natural language processing, and quantum computing.
Defensive approaches include adversarial training, differential privacy, data sanitization, anomaly detection, model integrity verification, provenance checking, input normalization, and checkpoint recovery.
The experiments reconstruct attacks from Book 2, apply corresponding defenses, measure performance, and then reevaluate each system against an adaptive adversary who knows the defense is present.
Importantly, the collection preserves unsuccessful and partially effective defenses to demonstrate the limitations of individual security controls.
Designed for AI security researchers, educators, and developers studying reproducible adversarial robustness experiments. These are educational research artifacts, not certified production defenses or evidence of a general quantum advantage.
Comments2
Rebuilding the Book 2 attacks first means a reader follows the same 35 systems from broken to protected. Is a Book 4 planned, and will you pitch the series to universities as a required text for security courses?
Great question, Ryan! At this point, there are no plans for a Book 4. The three-book series was designed to take readers through the complete process of building AI models, attacking them, and then defending them.
As for universities, absolutely! One of my goals is to make the series available as a hands-on educational resource for AI, machine learning, and cybersecurity courses. The Google Colab notebooks and reproducible experiments make it particularly suitable for classroom instruction without requiring extensive infrastructure.
I also see potential for using the series as the foundation for professional certificate programs. Appreciate your interest and suggestions!
2 comments
Rebuilding the Book 2 attacks first means a reader follows the same 35 systems from broken to protected. Is a Book 4 planned, and will you pitch the series to universities as a required text for security courses?
Great question, Ryan! At this point, there are no plans for a Book 4. The three-book series was designed to take readers through the complete process of building AI models, attacking them, and then defending them.
As for universities, absolutely! One of my goals is to make the series available as a hands-on educational resource for AI, machine learning, and cybersecurity courses. The Google Colab notebooks and reproducible experiments make it particularly suitable for classroom instruction without requiring extensive infrastructure.
I also see potential for using the series as the foundation for professional certificate programs. Appreciate your interest and suggestions!