The Schwartz Report

Blog archive

Machine Learning Is Key To Improving Accuracy of Next Microsoft Band

When Microsoft came out with its Fitbit-like band last year, it introduced some interesting new capabilities to the crowded market for such gadgets. But one of the reasons I returned the first iteration of the Microsoft Band after using it for a month was that it didn't appear to render precise and consistent heart rate data. I have come to learn I wasn't the only one to come to that conclusion and decided to wait and see what the next version offers before spending $200. In all fairness, not all share that view.

It appears the company is readying the next version for the upcoming fourth quarter holiday season, according to published reports. In a briefing at the Microsoft Research Center on the Redmond campus yesterday, Corporate VP Peter Lee, who oversees new experiences at the lab (NExT), indicated that the next Microsoft Band will render more accurate heart rates thanks to improved sensors, though he didn't get into the timing of the release. These improved sensors aren't coming in the form of better hardware but rather major advances in the software developed by Microsoft, he explained.

These same advances apply to other work such as Microsoft's Bing platform, which is all emerging as major Microsoft assets for Microsoft's efforts to advance machine learning. Lee gave the brief discussion about the Microsoft Band as an example of a skunkworks project called "Jewel" that came out of its research labs focused on applying machine learning. While machine learning has long been a key focus at Microsoft Research, the company this year took a step forward with the release of Azure ML. The compute and storage enabled by cloud-based machine learning has helped improve the algorithms used to render data such as blood flow, according to Lee, speaking with journalists Thursday at Microsoft's research center on the company's Redmond campus.

Lee admitted that the hardware BOM (bill of materials) included in the sensors of the Microsoft Band was limited and indicated that won't change in the next version. "We found the signal was woefully inadequate, especially when under physical stress, the key use-case for the Microsoft Band," he said. "While I am making disparaging remarks, it is on par or better than what you would find in the Apple Watch and other fitness bands."

It'll be interesting to see how the improvements in the software algorithm contribute to the next version of the Microsoft Band both in terms of accuracy and other yet undisclosed features of the new device.

Posted by Jeffrey Schwartz on 08/28/2015 at 12:32 PM


comments powered by Disqus

Subscribe on YouTube