Learning in Situ - A Randomized Experiment in Video Streaming

Explore the findings of a large-scale randomized controlled trial on video-streaming algorithms for bitrate selection and network prediction. Dive into the challenges faced by sophisticated and machine-learned control schemes in real-world settings, and discover why a simple buffer-based control outperformed them. Examine the statistical analysis revealing the impact of heavy-tailed network and user behavior on algorithm performance. Learn about a novel ABR algorithm that leverages deployment data and supervised learning in situ to achieve robust performance. Gain insights into the development of probabilistic predictors for chunk transmission times and their integration with classical control policies. Access the weekly published archive of data and results, and learn how to participate in this ongoing study to advance research in bitrate selection, network prediction, and congestion control for video streaming.