ENHANCING GENERALIZATION FOR NEURAL ADAPTIVE VIDEO STREAMING USING REPTILE META-LEARNING

Adaptive video streaming is of major significance in this new digital era for transportation of high-quality multimedia data on time-varying networks.In this paper, we suggest a lightweight meta-RL framework to enhance the generalization and online adaptation skills for bitrate choice on the basis of Reptile.The model can instruct a universal initialization on a broad variety of network situations, and adapts rapidly in-stream to maximize video quality and decrease stalling, In comparison to classical ABR and, the suggested method can perform a faster adaptation at lower meta-critic and complexity systems and thus is suited for contemporary, scalable video streaming platforms.