Microsoft has rolled out improvements to its facial recognition system in an effort to better identify the gender of individuals with darker skin tones using artificial intelligence.
It's no secret that facial recognition technologies are used everywhere for various applications ranging from unlocking mobile devices to recognizing crime suspects. However, current biases can limit the system's capabilities. Microsoft noted in a blog post that existing commercial facial recognition technologies have higher error rates in recognizing women with darker skin than in identifying men with lighter skin tones.
The software giant attributed that limitation to a narrow training dataset. To address that challenge, Microsoft's facial recognition team worked to improve its gender classifier system by removing biases on skin tone. Specifically, the team collected more data, improved the precision of the classifier, and refined its data collection efforts with a particular focus on skin tone, gender, and age.
The new improvements led to a significant reduction in error rates: up to 20 times for men and women with darker skin and nine times for all women. Microsoft's facial recognition system is available as the Face API via Azure Cognitive Services, which recently gained improvements with the addition of search and vision capabilities.