Intelligent Windows Phone Microsoft's big investment in machine learning
Microsoft Research's biggest investment is machine learning, the R&D division's new chief has revealed, claiming the team is "well within reach of solving speech recognition" as well as intelligently analyzing the content of images. "Machine learning is the really big one" Peter Lee told MIT Technology Review, "it is our number one investment." With the Nokia acquisition in the pipeline, Lee also says that the goal is to make tomorrow's Windows Phones "aware of what you are doing" through sensors and wearables.
"We are committed to making sure that the best concepts in hardware, devices, and sensors are in phones" the new head of research said, pointing out that Windows Phone began as a Microsoft Research-led skunkworks project operated with a sub-team of Microsoft's phone business. In fact, he argues, the first examples of that are already embedded in the platform, such as the learning keyboard which invisibly tracks where users' fingers commonly hit the on-screen buttons and shape the auto-correction and word prediction accordingly.
However, Lee suggests, even more intelligent smartphones are in the pipeline. "We think we're well within reach of solving speech recognition, making a big dent in translation, and devices that see and hear with capabilities approaching human ones" he points out.
That might one day mean our phones or wearables like Google Glass not only snap photos of our surroundings, but can themselves understand what's going on. "So, for example, a camera could understand what is being said and what it is looking at" Lee explains. "A photo could include this extra information. Or [a phone] could look at a plate of food and understand what it is, to help your diet and monitor your health."
Basic attempts at machine learning of that sort have been tried before. Google Goggles, for instance, can identify certain landmarks or objects in images, while previous research from the search giant has turned supercomputers into cat-spotters. NVIDIA resurrected the test itself, with a GPU-powered neural network earlier this year.
That's a long way from the same sort of power inside a phone or tablet, however. The Nokia acquisition isn't final, Lee points out, though the Finnish firm is known to be working on its own wearables and distributed computing system, and its own R&D teams are likely to fall under the Microsoft Research umbrella.