Alloy: A Fusing Algorithm for Adaptive Video Streaming with Meta-Reinforcement Learning

Abstract

In this work, we propose Alloy, an ABR algorithm that fuses existing algorithms to cover a wide variety of environments. Alloy provides high QoE to all users by combining the strengths of various algorithms. To do this, we find transferable parameters between multiple algorithms via meta-updates. Finally, with transferable parameters, Alloy covers a wide range of environments, providing different policies in different environments with just a few update steps.