To demonstrate the new machine learning capabilities in its Azure cloud, Microsoft has set up a Web site that guesses the ages of people in uploaded photos.
The site, How Old Do I Look, is meant to show how easy it is for developers to use machine learning algorithms to make predictions and spot trends, said Joseph Sirosh, Microsoft corporate vice president in the company’s cloud and enterprise group.
The site, which he unveiled on Thursday at Microsoft’s Build 2015 developer conference in San Francisco, asks the user to upload a picture with someone’s face. After a minute or two, it will return a guess for how old the person in the photo is.
Behind the scenes, the demonstration uses the Azure machine learning service, first to identify faces in a photo and then to guess people’s ages.
Machine learning is a type of data analysis that allows computers to draw inferences from large sets of data, by building predictive models through repeated sampling of data. Because machine learning typically requires a large amount of computing power, its use has mostly been restricted to the academic community until fairly recently.
Thanks to cloud computing, machine learning is now becoming more affordable for businesses. Microsoft launched its commercial machine learning service in February.
Thursday’s demo at Build was designed to familiarize developers with how machine learning could be used, and embedded into applications. Sirosh used a picture of a family whose members ranged in age from the teens to the early 70s. The service was able to predict their ages fairly closely, he said. The service can even guess the age of people depicted in paintings. It calculated the Mona Lisa’s age as 23, which was about the age of the model in the Leonardo da Vinci painting.
Machine learning is an iterative process, so the more data the system gets, the more accurate its predictions can be, Sirosh said during the keynote speech. Many predictions won’t always be correct. In one test done by IDG News Service, How Old Do I Look examined a recent low-resolution photo of Microsoft CEO Satya Nadella, and concluded he was 59, 12 years older than his actual age of 47.
The demo also illustrated how the Azure Streaming Analytics service can gather metrics about Web site usage in real time. Sirosh asked the audience, and those watching the keynote by webcast, to try the demo. He then went to the Streaming Analytics page, which provided numerical summaries of those using the service, the genders and ages of those in the photos, as well as a map showing where these users were located in the world. The page showed the bump in usage almost immediately.
Shirosh also spoke of other, more industrial uses, of Azure machine learning. For instance, Fujitsu has built a system using Azure to help Japanese farmers predict the best time to get dairy cows impregnated after they go into heat, which would help them breed the cows more efficiently.
Each cow is equipped with a sensor on one of its legs. Like a bovine Fitbit, the device records how many steps each cow takes in a given day. Females cows who are about to go into heat tend to walk around a lot more than usual. The sensor data is transmitted to Azure, which then can identify the perambulating cows, and send alerts back to farmer identifying those that might be in heat.
The age old method of detecting a fertile cow, by close observation, has only been 33 percent, but the automated procedure can bump that up to 95 percent, Sirosh said.