Reinforcement Learning Based Adaptive Video Streaming on Named Data Networking

Abstract

© 2020 IEEE. Under complex network conditions, adaptive video streaming requires additional state information for optimal quality selection. In this paper, we present the applicability of reinforcement learning techniques on NDN adaptive streaming. Both buffer-based and throughput-based adaptation are studied and observed their characteristics. The Q-learning algorithm is used to learn state-action values. Based on a greedy policy, the simulation results demonstrate that RL agents tend to choose the best possible bitrate which consequently reduces the quality fluctuation in adaptive streaming.