MUMBAI, India, Aug. 10 -- Intellectual Property India has published a patent application (202641089937 A) filed by Vellore Institute Of Technology on July 24, 2026, for System And Method For Privacy-Preserving In-Vehicle Recommendation Using Federated Learning And Cross-Domain Feature Alignment.
Inventors include Ashok B; and Vatsal Jha.
The application for the patent was published on July 31, 2026, under issue no. 31/2026.
Abstract: ABSTRACT The present invention relates to a privacy-preserving in-vehicle recommendation system based on behavior-clustered federated learning for heterogeneous connected vehicle environments. The system acquires telemetry from GPS and OBD-II sensors, synchronizes and normalizes the sensor data, and generates behavior embeddings representative of driving patterns. A tier-adaptive feature imputation mechanism reconstructs unavailable sensor features using hardware availability masks, enabling unified model training across vehicles with different sensor configurations. The generated behavior embeddings are protected using client-side Local Differential Privacy before transmission to a federated server, which clusters vehicles according to semantic driving behavior and distributes cluster-specific recommendation models. A cross-domain contrastive alignment mechanism associates driving behavior with destination and content preferences to enable personalized recommendations without requiring overlapping users or historical ratings. A behavior-driven cold-start routing mechanism assigns newly participating vehicles to appropriate behavioral clusters, thereby providing immediate personalized recommendations while preserving user privacy, accommodating heterogeneous hardware, and improving recommendation accuracy in intelligent transportation systems. (Fig. 1)
Disclaimer: Curated by HT Syndication.
