Researchers at MIT think that AI can solve Video Buffering issue

A new Video Delivery Algorithm that wouldn’t take entirely too full bandwidth. So before, that information is cut up into shorter bits and transmitted sequentially. But to guarantee that the video quality is adequate, sites like YouTube support ABR (Adaptive BitRate) algorithms to decide at what resolution the video will play. ABRs usually come in two styles: those that include how fast an interface can transmit data and that work to keep a sufficient buffer at the top of the video.

If the rate-based algorithm breaks, the video will experience pixelation as the method drops the bitrate to guarantee that the video keeps playing. But if thou try to skip too far ahead, it creates havoc with the buffer-based system which later has to freeze playback while it fills both the new chunk of video and the buffer before of it. Both of those ABRs are actually addressing two sides of the same overarching issue but not is fully able of solving it. And that’s where AI comes in.

There’s really now been a bit of investigation into this issue. A research team from Carnegie Mellon lately developed a “model predictive control” (MPC) system that tries to predict how network provisions will change over time and perform optimizations choices based on that model. The difficulty with that system, though, is that it will only ever be as great as the model itself and it is ill-suited for systems that see sudden or drastic variations in traffic flows.

CSAIL’s AI, dubbed Pensive, seems not rely on a model. Instead, it’s used machine learning to calculate out when and under what conditions to switch in rate and buffer-based ABRs. Like other neural networks, Pensive practices rewards and penalties to weigh the outcomes of each trial. Over time, the method is able to attune its behavior to consistently gain the highest reward. Interestingly, since the awards can be adjusted, the whole system can be attuned to behave however we want.