CertifiedRobustnessReport<T>
Models & Types · AiDotNet.AdversarialRobustness
The certified-robustness audit produced by a configured certified defense, plus — when an empirical attack is also configured — the certified-versus-empirical robustness sandwich.
How It Works
Certified accuracy is a guaranteed lower bound on robustness (proven safe within a radius); empirical robust accuracy from an attack is an upper bound (only “no attack found yet”). The true robustness lies between them. Reporting both at the same perturbation budget — and the gap between — is beyond tools that give one bound or the other.
Constructors
| Constructor | Summary |
|---|---|
CertifiedRobustnessReport() |
Properties
| Property | Summary |
|---|---|
CertifiedAccuracyMetrics<T> Metrics { get; set; } |
The certified accuracy metrics (certified accuracy, certification rate, mean/median radius). |
bool SandwichAvailable { get; set; } |
Whether the certified-vs-empirical sandwich was computed (both bounds available). |
double RadiusUsed { get; set; } |
The perturbation radius the certification was evaluated at. |
double? CertifiedEmpiricalGap { get; set; } |
The proven-vs-unproven gap: empirical robust accuracy minus certified accuracy, when both are available. |
double? EmpiricalRobustAccuracy { get; set; } |
The empirical robust accuracy at the same radius (from the configured attack), or null when no attack was configured — the upper bound of the sandwich. |
string DefenseName { get; set; } |
The configured defense’s type name. |