How Machine Learning can put an end to Video Buffering

Everyone hates watching their YouTube video buffer. We’ve been there with you, languishing in impatience.

Everyone also knows that machine learning is the basis for developing an artificial intelligence driven future right out of science fiction.

So what does machine learning have to do with video buffering?

New Research from MIT Takes Aim at Video Buffering

Machine learning focuses on learning abilities. You’re actually teaching machines how to learn things which, again, is the basis for a theatrical AI future. So how did MIT utilize machine learning to effectively eliminate video buffering? Simple: something called neural adaptive video streaming with the Pensieve System.

MIT researchers Hongzi Mao, Ravi Netravali, and Mohammad Alizadeh collaborated to develop this AI. “Pensieve”, aside from being an object used for memory viewing from Harry Potter, is a word that seems to be the combination of “pensive” and “sieve”, and it works simply and effectively. The AI uses an adaptive bitrate algorithm to discern which video quality works best on your network while avoiding any buffering breaks in the video. Due to current ABR algorithms, the researchers saw an opportunity.

As stated in the research paper, many current ABR algorithms “use fixed control rules based on simplified or inaccurate models of the deployment environment”. This leads to subpar results or total failure in achieving a high quality of experience. But with machine learning, an AI can out perform your standard ABR algorithm.

No More Trade-Offs

Companies like YouTube (read: Google) and Netflix already try to mitigate video buffering. Due to constraints, they can only mitigate so much. If net neutrality disappears, those constraints could grow considerably. 

As a result, they have to sacrifice pre-buffering the next video segment for the quality of the video or vice versa. The Computer Science and Artificial Intelligence Laboratory (CSAIL) at MIT researched Pensieve to make that tradeoff seamless.

As a result of teaching the AI to favor certain conditions based on network, location, the volume of users, etc, Pensieve can cut rebuffering rates up to 30%. Of course, they have only been able to tinker with about a month’s worth of content. If the Pensieve team could utilize other streaming platforms like Hulu or the entire catalog of Netflix, what else could the AI learn?

Other Uses for a Video Buffer Eliminating AI

Not only is the Pensieve useful for your average video streamer, it could be instrumental in the future of high-resolution VR content streaming.

In a world of autonomous cars, AI assistants, and reusable rockets, no one should have to wait for a YouTube video to buffer. No one! 

What other everyday activities could machine learning improve?