MUMBAI, India, May 1 -- Intellectual Property India has published a patent application (202621028093 A) filed by Mrs. Swati Shailesh Jadhav; and Dr. Jyoti Yadav, Pune, Maharashtra, on March 10, for 'leucocephalus social swoop optimization method and system for reliable client updates in federated learning.'
Inventor(s) include Mrs. Swati Shailesh Jadhav; and Dr. Jyoti Yadav.
The application for the patent was published on May 1, under issue no. 18/2026.
According to the abstract released by the Intellectual Property India: "Leucocephalus Social Swoop Optimization Method and System for Reliable Client Updates in Federated Learning This relates to a system (100) and method (300) wherein, the server (104) receives local model updates generated by the user devices using data stored locally at the respective devices. The optimization comprises evaluating the local model updates, organizing the user devices into one or more groups using a predefined grouping mechanism including randomized assignment, modifying the updates based on interactions among the updates and one or more reference updates, and selectively retaining modified updates based on an improvement condition. The retained local model updates are further refined using population-level guidance derived from a reference update and an aggregate population trend. Reliability metrics are computed for the refined local model updates based on deviation analysis relative to one or more reference statistics. Accordingly, a subset of the refined local model updates aggregated with at least one global reference update to update a global federated learning model enhancing robustness and resilience against unreliable or adversarial client behavior."
Disclaimer: Curated by HT Syndication.
