When Wireless Video Streaming Meets AI: A Deep Learning Approach
When Wireless Video Streaming Meets AI: A Deep Learning Approach
复制标题
DOI：
10.1109/mwc.001.1900220
复制
发表时间：
2020-04
影响因子：
12.9
通讯作者：
Lu Liu;Han Hu;Yong Luo;Yonggang Wen
中科院分区：
文献类型：
--
作者：
Lu Liu;Han Hu;Yong Luo;Yonggang Wen
无线多媒体大数据包含了用户行为、内容特征和网络动态等有价值的信息，可以驱动系统设计和优化。如何挖掘数据智能并将其融入无线多媒体系统是无线多媒体系统的根本问题。受深度学习成功的启发，在这项工作中，我们提出并展示了无线多媒体系统和深度学习的集成。我们首先将无线多媒体系统分解为三个组件，包括最终用户，网络环境和服务器，并提出了几个潜在的主题来拥抱深度学习技术。然后，我们提出了基于深度学习的QoS/QoE预测和比特率调整作为两个案例研究。在前一种情况下，我们提出了一个端到端的统一框架，包括三个阶段，包括数据预处理，表示学习和预测。与最佳基线算法相比，它实现了显着的性能改进（88%对80%）。在后一种情况下，我们提出了一个基于深度强化学习的比特率调整框架。评估性能与真实的无线数据集，我们表明，感知视频QoE平均比特率，重新缓冲时间和比特率变化可以显着改善。
Wireless multimedia big data contains valuable information on users' behavior, content characteristics and network dynamics, which can drive system design and optimization. The fundamental issue is how to mine data intelligence and further incorporate them into wireless multimedia systems. Motivated by the success of deep learning, in this work we propose and present an integration of wireless multimedia systems and deep learning. We start with decomposing a wireless multimedia system into three components, including end-users, network environment, and servers, and present several potential topics to embrace deep learning techniques. After that, we present deep learning based QoS/QoE prediction and bitrate adjustment as two case-studies. In the former case, we present an end-to-end and unified framework that consists of three phases, including data preprocessing, representation learning, and prediction. It achieves significant performance improvement in comparison to the best baseline algorithm (88 percent vs. 80 percent). In the latter case, we present a deep reinforcement learning based framework for bitrate adjustment. Evaluating the performance with a real wireless dataset, we show that the perceived video QoE average bitrate, rebuffering time and bitrate variation can be improved significantly.