Application of reinforcement learning in dynamic vidieo streaming

Abstract

Video streaming is the most popular service all around the world. On an annual report in 2019, Sandvine, a leading provider of network intelligence solutions, stated that more than 60% of traffic on the internet is originated from video streaming services (Sandvine, 2019). This figure, however, is expected to increase even more in the upcoming years due to the transition from HD video bitrate to UHD or 4K video streaming as the number of 4K TV connection increase according to Cisco report (Cisco, 2020). This explosion in popularity is an opportunity but also raises many questions for service providers. One of which is the challenge of maintain high quality of experience (QoE) for customer across a massive range of device, network condition and bitrate quality. This is especially challenging due to customer desired to have high video bitrate while capable of avoiding lag and re-buffering which are, in general, conflict goals. In this thesis I would like to have a look at a popular solution for the prior problem as well as discussing about the various methods that being used or develop that going with the solution