MUMBAI, India, Aug. 24 -- Intellectual Property India has published a patent application (202641097529 A) filed by K. Harisudha; G. Elavel Visuvanathan; and S. Murugaveni on August 12, 2026, for Secure Federated Learning With Encrypted Aggregation And Communication-Efficient Distributed Model Training.
Inventors include K. Harisudha; G. Elavel Visuvanathan; and S. Murugaveni.
The application for the patent was published on August 21, 2026, under issue no. 34/2026.
Abstract: ABSTRACT SECURE FEDERATED LEARNING WITH ENCRYPTED AGGREGATION AND COMMUNICATION-EFFICIENT DISTRIBUTED MODEL TRAINING A privacy-preserving communication-efficient federated learning framework is disclosed for distributed learning of a distributed machine learning model. A shared model is trained on the locally available data by each client and a model update is created. A communication-efficiency controller modifies the update based on the network and device characteristics by selecting the parameters, sparsifying the update, quantizing and/or compressing it. The updated version is cryptographically secured prior to sending. An aggregation module that performs aggregation using protected updates without revealing individual plaintext updates to the aggregation module is called an encrypted aggregation module. The recovery and application of an authorized aggregate update are applied to the global model. Bandwidth, latency, device resources, training convergence, etc., can be dynamically controlled for clients and for the parameters of the update transmission. The framework thus lowers communication cost and training delay, enhances the confidentiality of distributed model updates and allows scalable federated training of artificial-intelligence models across heterogeneous networks.
Disclaimer: Curated by HT Syndication.
