We focus on predictive modeling to enable proactive resource management in cloud and network systems. It leverages advanced learning-based algorithms, multi-feature fusion techniques, and cloud–edge collaborative strategies to anticipate future workloads and network conditions. These predictions improve the foresight of resource allocation, enhancing overall system efficiency and service quality. Our research group has been actively exploring predicting key future trends – from data center workloads to content delivery and network performance – to guide proactive actions. Its scope spans multiple domains, including content distribution, cloud-edge workload management, network performance optimization, and device-to-device (D2D) content propagation. By providing accurate foresight, this research helps minimize latency, prevent resource contention, and improve users’ quality of experience through timely, informed resource provisioning. (1) For practical contributions, we have delivered fine-grained demand forecasting for 5G content caching, unified workload prediction for multi-tenant edge clouds, latency-aware network optimization, and D2D propagation prediction—collectively enabling proactive resource allocation, higher utilization, and lower latency across real deployments. (2) For theoretical contributions, we have proposed a multi-feature GCN-LSTM for spatio-temporal forecasting, a one-for-all Transformer-based framework (DynEformer) for dynamic workloads, and a dual-granularity, reconciliation-guided model (DURABLE) that enforces hierarchical coherence while capturing spatial-temporal heterogeneity, advancing the state of predictive modeling in distributed computing systems.
Our work in Computation Forecasting has generated high-impact publications and practical, deployable solutions that address key challenges in edge intelligence and resource management—accelerating the transition of advanced predictive technologies from the lab to large-scale, real-world systems.
