Machine Learning based Bandwidth Prediction for Dynamic Adaptive Streaming over HTTP
Soyoung Yoo, Gyeongryeong Kim, Minji Kim, Yeonjin Kim, Soeun Park, Dongho Kim · JOURNAL OF ADVANCED INFORMATION TECHNOLOGY AND CONVERGENCE · 2020
By Digital Transformation, new technologies like ML (Machine Learning), Big Data, Cloud, VR/AR are being used to video streaming technology. We choose ML to provide optimal QoE (Quality of Experience) in various network conditions. In other words, ML helps DASH in providing non-stopping video streaming. In DASH, the source video is segmented into short duration chunks of 2-10 seconds, each of which is encoded at several different bitrate levels and resolutions. We built and compared the performances of five prototypes after applying five different machine learning algorithms to DASH. The prototype consists of a dash.js, a video processing server, web servers, data sets, and five machine learning models.