In recent years, Satellite-Terrestrial Networks (STNs) have garnered significant attention for extending network coverage to areas beyond the reach of traditional terrestrial networks. With the rapid expansion of STN applications, integrating Service Function Chaining (SFC) technology has become crucial for delivering differentiated services. However, the dynamic and complex structure of STNs presents significant challenges for SFC deployment. To address these, we propose a multimodal reinforcement learning algorithm that uses separate neural networks to process diverse STN data, enabling more effective SFC deployment decisions. Our approach includes a Graph Transformer for processing network states represented as graphs, capturing the relationships between nodes, links, and resource distributions. Additionally, two MLPs are used to handle QoS requests and global network information. Built on these components, the Proximal Policy Optimization (PPO)-based algorithm demonstrates superior performance over conventional AI methods, effectively learning optimal SFC deployment strategies.

Federated Learning (FL) is an emerging distributed learning paradigm that enables collaborative model training among multiple clients without requiring access to their private local data. However, in practical settings, FL commonly encounters both model heterogeneity and data heterogeneity. Variations in network architectures and patterns of non-independent and non-identically distributed (non-IID) data across clients often lead to significant degradation in global model performance. Existing approaches that utilize knowledge distillation to address heterogeneity in FL typically rely on access to public datasets or partial sharing of private data. Such reliance increases communication overhead while introducing privacy concerns, which impede practical deployment. In this work, we propose a novel heterogeneous federated learning framework, referred to as FedPAG. Our method extracts statistical information from Batch Normalization layers to synthesize pseudo-data aligned with local distributions, enabling similarity-aware aggregation of student models on the server while circumventing the need for external or shared data. Meanwhile, each client preserves and transfers its local knowledge via local knowledge distillation, enabling personalized optimization. By eliminating the dependence on public datasets and reducing privacy risk, FedPAG provides a scalable and privacy-preserving solution for heterogeneous FL. Rigorous experiments on three real-world datasets under diverse non-IID settings validate FedPAG’s superior performance and robustness over baseline methods in heterogeneous FL environments.

The integration of Software-Defined Networking (SDN) and Artificial Intelligence (AI) presents promising opportunities for managing and optimizing LEO satellite network routing. However, as the scale and coverage of satellite networks continue to expand, challenges are posed to both centralized and distributed architectures in terms of managing network information and coping with routing complexity. To overcome these challenges, leveraging distributed SDN technology, a stigmergy multi-agent hierarchical deep reinforcement learning routing algorithm is proposed in multi-domain collaborative satellite networks. A pheromone-based mechanism is incorporated to facilitate collaboration during independent training, and hierarchical control is employed to decouple the complexity of cross-domain routing decisions. Simulation results demonstrate that our proposed algorithm exhibits good scalability and performance in large-scale satellite networks.