MUMBAI, India, Aug. 31 -- Intellectual Property India has published a patent application (202641101849 A) filed by Vellore Institute Of Technology on August 24, 2026, for Privacy-Preserving Federated Learning System Using Context-Matched No- Event Sequences And Releasability-Certified Event Capsules.
Inventors include Shynu P G; Angel Elizabeth M; and D Jyotsna.
The application for the patent was published on August 28, 2026, under issue no. 35/2026.
Abstract: A privacy-preserving federated learning system (110) for remote monitoring includes a processor (114) and a memory (116) storing instructions executable by the processor (114) to receive multimodal observation data associated with a monitored entity from monitoring clients (102, 104, 106) and detect a candidate event. The processor (114) generates a context-matched no-event sequence based on a local history and a current context, encodes the observation data and the context-matched no-event sequence into respective latent representations, and determines an event residual based on a difference between the latent representations. The processor (114) decomposes the event residual into a locally retained private component and a candidate shareable component, generates a candidate event capsule, and determines privacy-risk, event-utility, and privacy-accounting measures. Upon failure of a releasability condition, the processor (114) transforms and re-evaluates the candidate event capsule. Upon satisfying the releasability condition, a releasability certificate is generated and transmitted for verification and certified aggregation to update a global event-prototype bank. FIG. 1
Disclaimer: Curated by HT Syndication.
