ML solutions for the real world problems of identity verification
summary
Biometric verification works but spoofing remains the hard problem: face encoders give tight feature vectors and top models hit ~99.8% on hard datasets, yet printed photos, screen replays, contact‑lens tricks and silicone masks still fool systems. The practical solution is a liveness pipeline of specialized ML classifiers trained on thousands of attack‑type samples, plus a feature matcher; building that data drives the effort (around 250 employees with ~16 dedicated to data creation) and enables flows like two minutes to open an account via selfie+ID+OCR.
AI-written summary · about
chapters (6)
- 0:00Introduction to biometric verification
- 1:33Three methods of identity verification
- 6:28Biometrics as the most convenient option
- 15:15How biometric verification works
- 20:54Fighting spoof attacks in biometrics
- 26:18Summary and future challenges
related talks
22mThe high cost of cheap biometrics
14mTransforming Digital Identity with AI
39mBeyond Recognition: A Cautionary Tale of AI and Identity Fraud
28mFace spoofing detection: theory and practice
08mAdvancements in facial recognition
20mInjection Attacks in Biometric
45mYour passport to the 7th continent
24mHow Biometrics Are Shaping the Future