Corsight AI, a facial recognition technology provider, has announced a reduction in both gender and racial bias within its Autonomous AI-powered solution, Fortify.
The results, published by the globally authoritative National Institute of Standards and Technology (NIST), highlight the organisation’s deeply-rooted commitment to eradicating biases within AI algorithms and follow shortly after various new capabilities were added to Corsight’s flagship product.
NIST Facial Recognition Vendor Test
The NIST Facial Recognition Vendor Test (FRVT) is widely recognised for setting standards within the industry and compares facial recognition algorithms from over 650 submissions. The bias analysis presented in the report shows the difference in the False Positive Rate (FPR) of gender and race compared to the FPR for white males.
Corsight’s newly published algorithm demonstrates excess FPR as low as 10% between white and black males
Corsight’s newly published algorithm demonstrates excess FPR as low as 10% between white and black males, standing out in comparison to the average excess of 130% for the top 30 leading VISA-Border submissions. Corsight’s algorithm also received top marks in the WILD, VISA-Border, and VISA-Kiosk tracks, which aim to simulate common use cases for facial recognition applications.
Reducing bias imbalance
Ran Vardimon, Ph.D., VP of Research at Corsight comments, “One of the main targets of Corsight’s Research and Development team is to reduce demographic bias, which in turn will enable improved identification that can be applied worldwide with the same accuracy. The FRVT report goes beyond measuring overall recognition accuracy by providing separate results for different demographics.”
“While most models suffer from a significant bias imbalance, analysis of Corsight’s recent submission demonstrates that both gender and racial bias has been reduced, leading to more reliable and accurate results. This illustrates the commitment that all of us at Corsight have to enable fair and ethical facial recognition.”
Fortify with advanced analytics
The update uses advanced analytics to calculate relevant statistics based on facial attributes
As part of its drive to eliminate bias and develop accurate solutions, Corsight recently launched the latest version of Fortify. This update uses advanced analytics to calculate relevant statistics based on facial attributes, without impacting privacy, while at the same time increasing speed and scaling capabilities.
Improving accuracy across age and gender has been front and centre of enhancements, enabling refined searches and better supporting streamlined and secure identity verification. This can improve a variety of use cases, including more accurate KYC (Know Your Customer) in Financial Services.
Accurate and privacy-focused facial recognition solutions
“Our mission at Corsight is to be the most ethical, accurate, fast performing, and privacy-focused facial recognition solutions provider on the market,” said Rob Watts, CEO at Corsight AI.
“Already our core product allows the recognition of faces in extreme angles, moving crowds, or low-quality images. Fortify can also detect faces with masks on and it has high accuracy rates in nearly complete darkness, at 2 lumens. We’re proud to have been recognised by NIST as making significant improvements where it matters the most. We will continue to ensure this technology is used ethically and as a force for good within society.”