Paravision Eliminates AI Bias in Multiple DHS Biometric Testing Categories

“In achieving zero-error performances across a range of age groups (and male and female genders), Paravision is not only doing its part to reduce AI bias in facial recognition, but demonstrating that such results can be achieved using the right engineering practices…”

Paravision has revealed the results of its performance in the Department of Homeland Security’s Biometric Technology Rally, laying claim to a remarkable zero percent error rate across multiple categories.

The company’s facial recognition technology achieved a 100 percent True Identification Rate on a range of demographics, and on images captured through three different acquisition devices. This included a 100 percent Matching system-focused True Identification Rate for both self-reported Male and Female groups; a 100 percent Matching-TIR across self-reported Asian, Black, and White racial categories; and a 100 percent Matching-TIR across Darker Skin and Lighter Skin demographic groups as measured using a calibrated color meter.

The latter performances in particular are noteworthy for addressing the racial bias issue that has come to plague the field of facial recognition. Multiple facial recognition systems have tended to perform with lower accuracy on non-white (and, often, non-male) demographic groups, an issue that can lead to racialized outcomes in real-world deployments. This is largely the result of a lack of diversity in the machine learning process, as facial recognition AI systems are often trained to recognize face biometrics on databases comprising mostly white, male faces.

In achieving zero-error performances across a range of age groups (and male and female genders), Paravision is not only doing its part to reduce AI bias in facial recognition, but demonstrating that such results can be achieved using the right engineering practices – a signal to the rest of the industry and to those who cite racial bias as a primary concern about the use of facial recognition in real world settings.

“Paravision relentlessly strives for outstanding face recognition performance across age, gender, race, and other characteristics, and it’s very rewarding to see that focus manifest in this level of operational performance,” commented Paravision CTO Charlie Rice. “Our work on accuracy improvements, specifically those inclusive of broad demographics, is never done. But 0% error rate across all groups and a range of capture devices in a test of this magnitude is a great milestone.”

“The data is solid and the numbers are clear: with the best products available in the market, face recognition can work for everyone,” added Paravision President and COO Benji Hutchinson.

The news about Paravision’s performance in the DHS’s testing program comes after the company delivered strong results in the National Institute of Standards and Technology’s Face Recognition Vendor Test in February. And it arrives after last month’s news that Paravision had established a Basic Ordering Agreement with the Department of Defense’s Joint Artificial Intelligence Center.

–

May 2, 2022 – by Alex Perala