Unidata Expanded Biometric Training Data
The firm increased demographic diversity in its liveness detection dataset to reduce algorithmic errors.
Updated on Oct. 14, 2026 in Cybersecurity
Live Poll
Does more diverse training data make you trust biometric liveness systems more?
Unidata has expanded its commercial biometric liveness detection training dataset to include 53,870 videos from 50 participants. This initiative aims to address uneven error distribution across different demographic groups in authentication systems.
Why it matters
Biometric liveness systems often exhibit uneven error rates across demographics, creating security and accessibility gaps. Research indicates that broadening training data and refining mask-matching techniques can significantly improve equity in these systems.
The dataset contains 53,870 videos featuring participants from African, Latin American, and Asian backgrounds. Research adjustments reduced accuracy variance between demographic groups from 3.07 percentage points to 0.75.
The players
Unidata
Unidata is a technology company specializing in commercial biometric liveness detection solutions.
Carnegie Mellon University Africa
Carnegie Mellon University Africa is a research institution based in Rwanda that focuses on engineering and technology development.
The details
Researchers at Carnegie Mellon University Africa developed methods to match visible skin areas to silicone mask tones, preventing system rejection based on color boundaries. Participants performed various presentation attacks using printed photos, phone replays, and custom masks created from plaster casts.
Timeline
The NIST evaluated passive biometric liveness detection algorithms during 2023.
The Tech Race
This development aligns with ongoing efforts to address findings from the 2023 NIST passive biometric liveness detection evaluation. It represents a pivot toward closing persistent accuracy gaps in identification technology that were highlighted by previous benchmarking studies.
Users may experience fewer authentication failures when interacting with biometric systems that have been trained on more diverse data. This improvement helps ensure that security measures are equally effective and accessible for all demographic groups.
The takeaway
Advancing biometric accuracy requires intentional training on diverse, representative datasets to correct historical performance imbalances. Developers should continue to implement rigorous testing against sophisticated presentation attacks to maintain system integrity.
Further reading
For more information on the current landscape of authentication standards, visit the Cybersecurity section.
Source note: This article includes information reported by Biometric Update.
Live Poll
Does more diverse training data make you trust biometric liveness systems more?
