MetaABR: Environment-Adaptive Video Streaming System with Meta-Reinforcement Learning

Abstract

This work focuses on a video bitrate algorithm that quickly adapts to new and various environments with just a few update steps. This aspect is especially important for large-scale video streaming services used by a wide variety of users in different environments. Our proposed model is based on a neural network and employs meta-reinforcement learning to train it. After training, it can be easily customized for a variety of new environments with a few update steps, providing a user-specific streaming service.