Federated Learning Node Model Selection and Sharing

The patent creates copies of local models trained on distributed data and stores them in a shared space accessible by multiple nodes. Instead of sharing sensitive raw data, nodes exchange and utilize model copies that encapsulate learned patterns, thereby maintaining data security while improving model accuracy through access to diverse data distributions across nodes.

The patent merges the storage and management of multiple local models into a shared space that is collectively maintained by multiple nodes. Instead of each node independently storing all local models (which would exponentially increase storage complexity), nodes share the burden of storing model copies, reducing individual device complexity while preserving model selection flexibility through the shared repository.

The patent enables information utilization by creating and sharing copies of local models across nodes, allowing efficient access to knowledge derived from distributed data. Privacy protection is maintained because only model copies (not raw sensitive data) are shared, and the shared space architecture controls access to these copies, thereby achieving both high information utilization efficiency and reliable privacy protection.

The patent segments the model management function by creating a distributed shared space where each node can independently store, access, and manage model copies. This eliminates the need for a centralized management system, allowing the system to scale distributedly while maintaining coordination through the shared space architecture. Each node retains autonomy to contribute and utilize models, enhancing both scalability and distributed adaptability.