Microsoft face-recognition tech gets some bias training

Face API via Azure Cognitive Services, the flagship facial-recognition technology at Microsoft, has been retrained to recognize a range of skin tones and genders. According to the announcement post, the upgrade reduces error rates for men and women with darker skin by up to 20%, with error rates for women reducing nine times. 

Working with experts on bias and fairness, Microsoft managed to model a gender classifier to get better results across all skin tones and recognise genders associated with search terms. The change mirrors similar steps taken by IBM in removing bias from facial recognition technology

"For example, if you do a web search for the word CEO, chances are you’ll get information about the senior leadership position in companies and organizations around the world, including a handful of images—most likely of men," said John Roach, editor of the AI blog at Microsoft. The results have more to do with SEO best practices that add necessary tags on images pertaining to the name, title, and company an image is associated with. It also has to do with the fact that under 5% of Fortune 500 CEO's are women.

Hanna Wallach, a senior researcher in Microsoft’s New York research lab has been looking into how best to surface search results which reflect gender-neutrality. Her team is rectifying the matter by incorporating diverse datasets in the development process in order to train systems in identifying a range of skin tones, gender and age groups. In doing so, Wallach is improving the classifier to produce higher precision results.