Email
Password
Remember meForgot password?
    Log in with Twitter

article imageMIT AI tries to eliminate buffering when streaming videos

By James Walker     Aug 14, 2017 in Technology
MIT has developed an AI capable of streaming video without making you wait for it to buffer. Buffering is one of the biggest hindrances publishers face when convincing viewers to watch their content. Visitors on slow connections will simply turn away.
Engadget reports that MIT's Computer Science and Artificial Intelligence Lab (CSAIL) has come up with a solution for buffering that could put an end to the loading icons and progress bars. It tasked a neural network with trying to solve one of the biggest challenges when developing conventional buffering systems.
"Buffering" refers to the chunk-based streaming of large files over a network. When you play a video, it's effectively chopped into sections that are downloaded individually of each other. You can start watching the video without having to download the entire thing first. If your connection is slow, you might end up watching a chunk before the next one has completely downloaded, creating a frustrating experience.
Modern buffering technology uses an Adaptive Bitrate (ABR) algorithm to adjust the quality of the video as it's played. If your network quality degrades while you're watching a video, most sites will automatically reduce its resolution so chunks continue to download quickly. ABR can be implemented in two ways, either by measuring the speed of a network and dynamically adjusting quality or always working to maintain a downloaded buffer at the end of the video.
READ NEXT: Emoji analysis helps AI detect sarcasm in tweets
The problem is that neither approach fully solves the issue. This is what CSAIL's new Pensive AI has finally overcome. It uses a combination of both ABR techniques, offering the benefits of each one. Current buffering systems use an algorithm to determine when to switch between the ABR mechanisms. Pensive has a neural network instead, allowing it to break free from rules and always favour smoother video playback.
Pensive gets smarter over time as it comes to understand how it can obtain its biggest reward. Like other neural networks, it determines success by looking for "rewards." In this instance, the ultimate goal is to always achieve smooth playback. Any stutter causes the AI to be penalised, forcing it to an alternative approach. As it learns over time, it comes to consistently look for the highest reward, always favouring smooth video and effectively eliminating stutter.
READ NEXT: Google's voice dictation tech now supports 30 more languages
According to the researchers behind Pensive, its behaviour could also be customised by the user. You'd be able to choose whether to allow some buffering and play at a higher resolution or to always force completely smooth playback.
"Our system is flexible for whatever you want to optimize it for," MIT professor Mohammad Alizadeh said in a statement to Engadget. "You could even imagine a user personalizing their own streaming experience based on whether they want to prioritize rebuffering versus resolution."
Pensive's success means it could eventually be adopted by commercial video streaming services. The team behind the AI first wants to explore its applications in another growing technology though. MIT is eyeing Pensive as a solution to smooth wireless virtual reality, allowing the headset cable to be removed. Pensive could overcome the hundred megabit bitrate required for VR, making wireless high-performance headsets a possibility.
More about Artificial intelligence, Ai, machine learning, neural network, Videos
 
Latest News
Top News