Future 6G applications demand ultra-low latency and high reliability, motivating the convergence of communication, computing, and multimedia services at the network edge. A central challenge is to align wireless network performance with user-centric quality of experience (QoE).
In this talk, I will present a data–model co-driven optimization framework that leverages imitation learning and the information bottleneck principle to enable adaptive video streaming under highly dynamic wireless conditions. The framework learns from offline expert policies, avoiding inefficient online exploration, and incorporates an adversarial information bottleneck to improve generalization. I will also introduce DeepBroadcast, a task-oriented semantic communication system for broadcasting. By combining semantic compression with dynamic fusion, DeepBroadcast adapts to diverse channel conditions and user requirements, significantly enhancing both efficiency and robustness. Guided by the information bottleneck framework, this approach highlights a unified path toward user-centric optimization in next-generation wireless networks.
In this talk, I will present a data–model co-driven optimization framework that leverages imitation learning and the information bottleneck principle to enable adaptive video streaming under highly dynamic wireless conditions. The framework learns from offline expert policies, avoiding inefficient online exploration, and incorporates an adversarial information bottleneck to improve generalization. I will also introduce DeepBroadcast, a task-oriented semantic communication system for broadcasting. By combining semantic compression with dynamic fusion, DeepBroadcast adapts to diverse channel conditions and user requirements, significantly enhancing both efficiency and robustness. Guided by the information bottleneck framework, this approach highlights a unified path toward user-centric optimization in next-generation wireless networks.
沪公网安备 31011502006855号