Microsoft is aiming to improve its users’ experience by enhancing a couple of its artificial intelligence (AI) assisted visual recognition offerings.
In a series of blog posts, the company announced its Face API had been updated to better recognise gender across skin tones, and that new intelligent visual search features have been added to the mobile app version of its Bing search engine. Both examples highlight Microsoft’s flourishing capabilities in artificial intelligence, which have lately been positively impacting the performance of almost all of its software. In other recent examples, several applications within Office 365 have received tweaks to improve in-app search results through selection by an intelligent algorithm.
The company admitted its facial recognition feature had fallen short of expectations for many users. It highlighted how the artificial intelligence industry as a whole must deal with certain challenges in ensuring visual recognition algorithms do not become biased by being fed only a limited training dataset.
Microsoft senior researcher Hanna Wallach commented on the difficulty in overcoming challenges to arise not from algorithmic or dataset failings but from the reality that human societies have their own biases. She said:
“If we are training machine learning systems to mimic decisions made in a biased society using data generated by that society, then those systems will necessarily reproduce its biases.”
Azure Cognitive Services program manager Cornelia Carapcea said her team had been working to provide “a more nuanced understanding of bias” and a stronger dataset had been needed to work across skin tones. Microsoft has said it is now able to reduce recognition error rates for people with darker skin tones by up to 20 times.
Of perhaps less sensitivity to many users, but still a worthwhile improvement, is the updating of Microsoft’s Bing mobile app to include an AI-driven visual search feature. The company has said it has compiled a vast dataset of tagged images to train its computer vision algorithm, which leverages the cloud to improve the speed of its calculations in real time. Bing Images product lead Vince Leung said:
“With the advent of cloud computing we are able to accelerate our ability to make sense out of pixels.”
For the end user, this means finding out information about almost any object is as simple as taking a picture of it. The company noted the difficulty of trying to find out information about plants and flowers in the wild is one it hopes Bing can now overcome, with Leung noting there are times when it is impossible to perform a search using words alone.
Microsoft also cited how shopping to a budget in the real world can be assisted by visual search. If a user were to spot a couch they liked in a store, they will be able to take its picture and find similar designs at a better price.
For more details on how the latest developments in technology can drive better performance at your business, get in touch with our friendly customer service team at WM Reply. We can provide the ideal collaboration solutions for you.