MUMBAI, India, Sept. 22 -- Intellectual Property India has published a patent application (202641105462 A) filed by Saveetha Institute Of Medical And Technical Sciences on September 02, 2026, for Privancy Preserving Federated Data Mining Framework Using Secure Gradient Aggregation And Adaptive.
Inventors include Thanupriya N; Dr. Sheeja Kumari; and Dr Ramya Mohan.
The application for the patent was published on September 18, 2026, under issue no. 38/2026.
Abstract: The invention discloses a privacy-preserving federated data mining framework that enables collaborative model training across multiple distributed data sources without requiring raw data sharing. The system incorporates a secure gradient aggregation mechanism that cmploys homomorphic encryption and noise-based differential privacy to protect intermediate model updates from unauthorized inference attacks. Each participating client device computes local gradients on its private dataset and transmits only encrypted or privacy-sanitized updates to a central or decentralized aggregator. To address the heterogeneity and asynchronous nature of client data, the framework further integrates an adaptive model synchronization algorithm that dynamically adjusts update frequency, model weighting, and synchronization intervals based on client performance, data distribution, and network conditions. This ensures stable convergence, improved model accuracy, and minimized privacy leakage during training. The proposed framework enhances data privacy, reduces computational overhead, and improves security against gradient-based reconstruction attacks, making it suitable for sensitive applications such as healthcare analytics, financial fraud detection, personalized services, and large-scale distributed data mining operations.
Disclaimer: Curated by HT Syndication.
