A Framework for Meta-Learning in Dynamic Adaptive Streaming over HTTP

Abstract

This work presents a framework with a taxonomy on meta-learning used in Dynamic Adaptive Streaming over HTTP (DASH). With the increasing complexity of network conditions and user preferences, there is a need for efficient adaptation mechanisms in DASH to provide optimal quality of experience (QoE) for users. Meta-learning, or learning to learn, has emerged as a promising approach to enhance adaptive streaming algorithms in DASH by leveraging prior knowledge and experiences. The proposed framework provides a systematic and structured approach for applying meta-learning techniques in the context of DASH. It encompasses essential components, including data collection and preprocessing, meta-model architecture, meta-training, meta-testing, fine-tuning, and continuous improvement. The taxonomy within the framework categorizes various aspects of meta-learning in DASH, such as meta-learning approaches, components, objectives, and applications