ML Security PapersPublished on arXiv
2604.25733
Input Manipulation Attack
OWASP ML Top 10 — ML01
Key Finding
Provides mathematical foundations and algorithmic approaches for formally verifying safety properties of neural networks in safety-critical applications
These lecture notes provide an introduction to the verification of neural networks from a theoretical perspective. We discuss feed-forward neural networks, recurrent neural networks, attention mechanisms, and transformers, together with specification languages and algorithmic verification techniques.
Key Contributions
- Comprehensive theoretical framework for neural network verification
- Coverage of feed-forward networks, RNNs, attention mechanisms, and transformers
- Specification languages and algorithmic verification techniques for safety-critical AI
🛡️ Threat Analysis
Verification of neural networks primarily concerns proving robustness properties and safety guarantees against adversarial inputs and edge cases - this is fundamentally about ensuring models behave correctly under input perturbations, which is the core concern of ML01 defenses.