Bias in Biometrics
summary
NIST FRVT testing shows measurable demographic variability in face recognition, so the concrete proposal is two pass criteria: an overall error threshold (example: <10% false-negative) and a variability limit (example: ±2% bound) evaluated with statistical confidence intervals, which drives necessary sample sizes across groups. Work includes educational videos on bias, published vendor-by-vendor FRVT demographic results (FRVT demographic results), evidence of variability in voice biometrics, and FIDO convening an ad hoc group to define testing and certification.
AI-written summary · about
chapters (8)
- 0:00Introduction to bias in biometrics
- 1:10Defining demographic bias in biometrics
- 3:19NIST findings on performance variability
- 6:14FIDO use cases and verification focus
- 9:51Educational video on bias and biometrics
- 13:28Testing and measuring demographic differentials
- 20:33FIDO’s role in bias testing and certification
- 21:36Q&A on voice and multimodal bias
related talks
27mEnsuring biometric recognition is secure and works well for all: A Case Study
27mThe Limitations and Danger of Facial Recognition
51mPrivate org & govt perspectives in navigating the tech landscape of remote digital identity proofing
43mMade to Measure? The Biases and Boundaries of Biometrics
06mFacial Recognition: Understanding Privacy Concerns and Attitudes Across Increasingly
01h 00mNavigating the Biometric Anatomy
24mHow Biometrics Are Shaping the Future
47mFace/Off: The Battle to Authenticate with Biometrics