MIT’s new artificial intelligence could kill buffering
For some, the sight of the buffer circle is enough to bring on spasms of existential angst. When that spinning circle of death appears, the digital world cracks, its illusory sense of control slips from your sweaty palm, and you are reminded, however briefly, that you are not the master of this realm, and you have no real idea how the machine you are using works. It’s also very annoying if you’re trying to show a video to someone.
Researchers at MIT may have come up with a way to stave of techno-existential panic for good, thanks to a new artificial intelligence system that can keep video steaming buttery smooth.
Buffering happens because video streaming occurs in chunks, with your device downloading sequential portions of a file that are then stitched together. This means you can start watching the video before downloading the entire thing, but if connection wavers you might finish one chunk before the next has been fully downloaded.
Sites like YouTube use Adaptive Bitrate (ABR) algorithms to work out what resolution a video should display at. In a nutshell, these allow the system to maintain the flow of images be measuring a network’s speed and lowering the resolution appropriately, or by working to maintain a sufficient buffer at the tip of the video. The issue is that neither of these techniques on their own can prevent annoying pauses in a clips if the network has a sudden drop in traffic flow – say, if you’re in a particularly crowded area, or if you’re moving in and out of tunnels.
MIT’s Computer Science and Artificial Intelligence Lab (CSAIL) AI, dubbed Pensive, takes these algorithms, but uses a neural network to intelligently work out when a system should flip between one and the other. The AI was trained on a month’s worth of video content, and was given reward and penalty conditions, to push it to calculate the most effective times to switch between ABR algorithms.
This system is adjustable, meaning it can be tweaked depending on what a content provider might want to prioritise – such as consistent image quality or smoother playback. “Our system is flexible for whatever you want to optimise it for,” commented MIT professor Mohammad Alizadeh in a statement. “You could even imagine a user personalising their own streaming experience based on whether they want to prioritise rebuffering versus resolution.”
While the death of the buffer symbol might be cause for celebration, the researchers also point to the benefits the AI system could have for virtual reality – potentially making it much easier for people to stream high-resolution VR games and films. “This is really just the first step in seeing what we can do,” noted Alizadeh.