Intelligent Service Intermediation via Federated Learning

The patent introduces federated learning as an intermediary mechanism that enables model training without direct data transmission. Local models are trained on edge devices using local data, and only model parameters (not raw data) are transmitted to the central server for aggregation. This intermediary approach allows the system to achieve model accuracy improvement while maintaining data privacy and security, as the central server never directly accesses raw user data.

The patent extracts only the essential model parameters from edge devices for transmission to the central server, rather than transmitting all raw data. This selective extraction of model updates (gradients, weights) allows global model training to be improved while significantly reducing transmission costs and network bandwidth consumption, as only compressed parameter information is transmitted instead of voluminous raw data.

The patent merges locally trained models from multiple edge devices through federated averaging, combining the knowledge gained from diverse local data sources. This merging process enables functionality optimization across disparate devices while maintaining data privacy, as the aggregation of model parameters from multiple devices creates a more versatile and adaptive global model without requiring centralization of raw data.

The patent segments the service processing architecture into distributed edge components and a lightweight central coordination layer. Local models process service interactions at the edge, maintaining simplicity and responsiveness, while the central server provides minimal coordination through model aggregation. This segmentation reduces system complexity and processing overhead compared to fully centralized processing, as computation is distributed and the central server only performs parameter aggregation rather than full service processing.