Optimizing adaptive video streaming with reinforced fine-tuning

With the prevalence of streaming video technology, the number of internet users enjoying video services is increasing. As a key algorithm for enhancing the Quality of Experience (QoE) in streaming video environments, adaptive bitrate (ABR) algorithms have witnessed significant development. However, traditional ABR algorithms suffer from insufficient adaptability and poor generalization, failing to deliver satisfactory QoE. To address these problems, this paper proposes a REFT-based ABR algorithm, which transforms QoE optimization into a two-stage problem. By leveraging REFT's multi-path exploration capability, the model's adaptability and generalization ability are effectively enhanced. Experimental results demonstrate a significant QoE improvement, with an average increase of approximately 45% over traditional ABR algorithms.