FABLE: federated adaptive belief learning for cloud-edge semantic caching
Year: 2026
Type: article
Abstract: Cloud-edge semantic caching must make proactive decisions from incomplete and evolving regional demand observations. We propose FABLE, a Federated Adaptive Belief Learning framework that maintains a posterior belief over latent demand and uses predicted belief trajectories to guide cache actions. A reliability-aware federated mechanism shares model knowledge across heterogeneous edge servers without uploading raw request traces, while local policy personalization preserves regional specificity. Under a non-stationary Noah-Wukong workload, FABLE converges faster than the baselines and reduces the last-50-episodes average latency by 8.69% relative to FDQN, with reductions of 29.37% and 73.31% relative to DDQN and WF-MCEC, respectively.
Authors: Yiming Zhang, Donglei Tao, Yunbi Wu, Shigui Li, Yi Ru
Institutions: China Tobacco, Zhejiang University
Language: English
Topic: Caching and Content Delivery
Subfield: Computer Networks and Communications
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W7223286185