Bamboo: Boosting Training Efficiency for Real-Time Video Streaming via Online Grouped Federated Transfer Learning

TL;DR

Bamboo is a novel online grouped federated transfer learning framework that significantly accelerates training for real-time video streaming, improving efficiency by up to 302% while maintaining QoE.

Contribution

The paper introduces Bamboo, a new framework that enhances online training speed for bitrate adaptation in real-time video streaming using federated transfer learning.

Findings

Training efficiency improved by up to 302%.

Maintains quality of experience (QoE) in diverse network conditions.

Outperforms existing reinforcement learning algorithms.

Abstract

Most of the learning-based algorithms for bitrate adaptation are limited to offline learning, which inevitably suffers from the simulation-to-reality gap. Online learning can better adapt to dynamic real-time communication scenes but still face the challenge of lengthy training convergence time. In this paper, we propose a novel online grouped federated transfer learning framework named Bamboo to accelerate training efficiency. The preliminary experiments validate that our method remarkably improves online training efficiency by up to 302% compared to other reinforcement learning algorithms in various network conditions while ensuring the quality of experience (QoE) of real-time video communication.