Citadel AI is excited to launch new Citadel Lens capabilities for validating biometric identification and verification systems.
Biometric systems that identify individuals, such as facial recognition with surveillance cameras or fingerprint-based identity verification, are strictly regulated by the EU AI Act and GDPR. Japan’s Act on the Protection of Personal Information also considers biometric information to be personal information and requires that it be handled accurately, fairly, and appropriately.
Citadel Lens can now automatically validate the quality of biometric systems against ISO/IEC 19795-1:2021 and ISO/IEC TR 29119-13:2022, for both identification (one-to-many matching) and verification (one-to-one matching) systems.
Beyond biometrics, Citadel Lens is utilized by a wide range of customers in the medical, automotive, financial, insurance, and other industries to test, monitor, and govern their AI systems, from generative AI to predictive AI, in a single platform.
Citadel Lens bridges engineering and governance teams by automatically generating technical reports for engineering teams who build AI systems, as well as governance reports for governance, risk, and compliance teams based on international and regional standards.
By evaluating biometric systems across multiple criteria such as accuracy, reliability, security, not only for the entire dataset, but also for specific data segments, you can measure technical performance as well as fairness and bias across populations.
Example evaluation criteria for biometric systems in Citadel Lens
| Metric | Definition | Objective |
|---|---|---|
| FTAR | Failure to Acquire Rate | Evaluates system reliability in an operational environment |
| FMR | False Match Rate | Evaluates the baseline risk of security breaches |
| FNMR | False Non-Match Rate | Evaluates the baseline risk of usability degradation |
| FAR | False Accept Rate | Evaluates the risk of security breaches in an operational environment |
| FRR | False Reject Rate | Evaluates the risk of usability degradation in an operational environment |
| FPIR | False Positive Identification Rate | Evaluates the risk of security breaches in one-to-many matching |
| FNIR per Rank | False Negative Identification Rate per Rank | Evaluates the risk of usability degradation and efficiency in one-to-many matching |
| Selectivity | Selectivity | Evaluates identification ability and efficiency in one-to-many matching |
To work with our team of AI experts and see how Citadel Lens can test, monitor, and govern your AI models and datasets, please contact us at any time.
