In announcing IDLive Face 1.18, the company did not disclose specific data concerning how the facial recognition system performs across subjects of different groups. But the company indicated that limiting bias in its biometric systems is “a top priority” for the company, and that a major focus of the latest platform upgrade was “closing accuracy gaps between ages, genders, and ethnicities.”
The move is aimed at addressing an increasingly widespread concern about the presence of demographic bias, and particularly racial bias, in AI systems – generally the result of training machine learning algorithms on datasets that are not sufficiently diverse. This can lead to serious real world outcomes such as false arrests and other symptoms of systemic racism.
As such, facial recognition vendors have become increasingly alert to the issue. In announcing its recent RealSense ID solution, for example, Intel emphasized that its biometric algorithm had been tested on a range of skin types and nationalities.
For its part, ID R&D says it has addressed the issue by implementing a “new machine learning pipeline” in its platform’s development.
That having been said, in offering his own commentary on version 1.18 of the solution, ID R&D President Alexey Khitrov emphasized the general advantages of IDLive Face in terms of overall performance, asserting, “unlike alternative liveness products we have had a relentless focus on enabling a zero-friction user experience. Improving speed and increasing accuracy, including both false accepts and false rejects, for all types of users and across various devices is critical to successfully using face for identity verification and authentication.”
ID R&D says that its number of IDLive Face customers grew exponentially in 2020 – a year in which COVID-19 drove unprecedented levels of traffic in digital channels – and that its platform is now conducting millions of passive liveness checks on a monthly basis.
January 19, 2020 – by Alex Perala