Edge Model Training for Efficient Cloud Updates

The patent divides the model training process into two segments: edge devices perform local training on aggregated data to create initial models, while the cloud server performs centralized training on combined data from multiple edge devices. This segmentation reduces the computing load on individual servers while maintaining high computing precision through distributed collaboration.

The patent introduces an intermediary mechanism where edge devices aggregate and pre-process data locally before transmitting to the cloud server. The edge devices act as intermediaries that perform initial training and model optimization locally, reducing the volume and complexity of data that needs to be processed by the cloud server, thereby maintaining precision while improving server efficiency.

The patent segments the centralized cloud-based system into a distributed architecture where edge devices autonomously perform data aggregation, local training, and model optimization. This segmentation enables each edge device to operate independently with its own data and models, eliminating dependency on continuous cloud connections while maintaining scalability through the distributed nature of the system.

The patent implements self-service capabilities at the edge device level, where each edge device autonomously aggregates its own data, performs local training, creates models, and shares them with the cloud server without requiring constant external intervention or reliable continuous connections. This self-service approach ensures system reliability even during intermittent or disconnected operations.