A wireless network client node individually protects each piece of a locally trained global ML model before transmitting them to a server node for aggregation.

Patentradars sammanfattning

This Ericsson PCT application (WO 2025/063910 A1) describes a method by which a client node in a wireless network individually applies distinct protection mechanisms to each of multiple pieces of a locally trained version of a global machine learning model prior to uploading them to a server node. The approach addresses privacy and integrity concerns inherent in federated learning deployments within 5G/next-generation networks. By disaggregating model updates into protected pieces rather than protecting the entire update as a monolith, the invention enables more granular security and potentially reduces vulnerability to inference or poisoning attacks on aggregated model updates.

AI-genererad och redaktionellt processad.