The end of buffering in streaming media?
We've all been there.
We settle down to watch our favourite YouTube video, and it starts, and then stops. And the dotted circle starts to spin.
That happens, of course, when there isn't enough bandwidth to stream a video, and so it has to build up a sufficient buffer before it begins to play again.
Services like YouTube and Netflix use a human-designed algorithm to determine what resolution to stream your content. But that algorithm frequently doesn't quite work, causing the video to pause while the content catches up.
Mohammad's new approach, at MIT's Computer Science and Artificial Intelligence Laboratory, is to use machine learning instead of an algorithm, a sophisticated AI called "Pensieve" that learns from thousands of streams simultaneously so it can adjust for any network speed or bandwidth restriction.
Unlike an strict algorithm, Pensieve doesn't set out with a prescribed set of rules to follow. It learns what works best on the fly, and then applies what it learns to improve streaming efficiency, he says. "We don't tell it what to do," he says. It observes and learns.
In tests done so far, Pensieve does up to 30 per cent better than the traditional algorithms used by streaming services, which is enough to eliminate buffering pauses completely in many videos.
While Mohammad says that Pensieve does require additional computing power, it's not a significant difference.
Perhaps most interesting, Pensieve isn't limited to just applying what it learns to streaming video. Mohammad says that ultimately, it could examine traffic in all sorts of networks to make them more efficient.
"There's a lot of potential for systems like this to really change how we design networks. Today, all of these things follow a similar design approach, where it's a human-engineered algorithm based on fairly simplified models. [AIs like Pensieve] could really improve experience for many different applications."