Researchers have developed a new end-to-end video protocol that would allow smoother and faster streaming of online videos.
More and more videos are being watched now than ever before. In the United States alone, around 85 percent of Internet users are reportedly watching video content monthly across different devices.
Further studies revealed that people’s preference for videos is not only limited to entertainment purposes. Many brands also depend on videos to boost their marketing efforts and sales. Hubspot reported that over 50 percent of consumers want more videos from their favorite brands or businesses.
However, the growing popularity of video content also led to video traffic congestion. This means people could experience excessive buffering and pixelation while watching online videos or streams.
But, researchers at MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL) wanted to give people a better viewing experience. That is by using a machine learning system that picks different algorithms depending on network conditions.
Machine Learning-Powered Video Protocol
Video streaming platforms like YouTube use adaptive bitrate (ABR) algorithms which break videos into small chunks that load as people watch. However, skipping to parts that are yet to be loaded would require the algorithms to stall and buffer those parts.
Now, if YouTube‘s ABR algorithms detected that a person’s Internet is slow, it will automatically adjust the video resolution. That is to ensure that the video will continue to play without interruptions. The only problem is that the images would be pixelated.
MIT researchers claimed that their machine learning-powered video protocol called Pensieve could stream videos with ten to 30 percent less rebuffering than current approaches. That’s also at levels rated by viewers ten to 25 percent higher on the quality of experience (QoE) metrics.
Lead author of the study, Ph.D. student Hongzi Mao, said:
“Our system is flexible for whatever you want to optimize it for. You could even imagine a user personalizing their own streaming experience based on whether they want to prioritize rebuffering versus resolution.”
How Pensieve Works
Pensieve doesn’t require a model or existing assumptions about things like network speed. Instead, it adjusts its algorithms using a rewards and penalties system.
“It learns how different strategies impact performance, and, by looking at actual past performance, it can improve its decision-making policies in a much more robust way,” Mao added.
Video streaming platforms like YouTube, Netflix, and Hulu could customize Pensieve’s system based on what metrics they want to prioritize for their viewers.
For instance, rebuffering at the beginning of the video is more acceptable to viewers. So, the system could be adjusted to give a penalty for rebuffering over time.
The video protocol was tested in different settings, including using Cafe Wi-Fi and LTE network while on the streets. The study revealed that Pensieve could still reduce rebuffering by ten to 30 percent under the said conditions.
Mao noted:
“When we tested Pensieve in a ‘boot camp’ setting with synthetic data, it figured out ABR algorithms that were robust enough for real networks. This sort of stress test shows that it can generalize well for new scenarios out in the real world.”
The team reported that Pensieve’s only been trained using a month’s worth of downloaded video content. They believe that training the system using large-scale data from video platforms could significantly improve its performance.
The team’s paper on their end-to-end video protocol is scheduled to be presented next week at the SIGCOMM conference in Los Angeles, California.
Comments (0)
Most Recent