Fast Adaptation of ABR Algorithm in Meta Learning Approach

Over the past years, the use of video streaming applications has surged significantly. There have been developments in adaptive bitrate (ABR) algorithms that use machine learning (ML) to enhance the quality of experience (QoE) for users. However, it remains uncertain if these algorithms maintain their effectiveness in today's intricate settings. Several meta-learning approaches have emerged, but the models still need a lot of updating to adapt to the environment. In this paper, we introduce a novel ABR algorithm designed to adapt to different environments while consistently delivering high QoE. By treating various environments as separate challenges, we manage to isolate ABR algorithm from direct environmental influences. In addition, we introduce online trainer and environment collector to further improve the adaptation ability in the online phase. We evaluation the system in a range of settings and confirmed its ability to adapt effectively to new and unforeseen environments.