MUMBAI, India, Aug. 12 -- Intellectual Property India has published a patent application (202641092318 A) filed by Vardhaman College Of Engineering on July 30, 2026, for A System And Method For Federated Machine Learning With Privacy-Preserving Data Processing.
Inventors include Ms. Uma Datta Amruthaluru; Ms. A Ashwini; Ms. Kalluri Niveditha; Mr. G E Agrawal; Dr. U Sesadri; and Prof. G Venkata Rami Reddy.
The application for the patent was published on August 07, 2026, under issue no. 32/2026.
Abstract: ABSTRACT A System and Method for Federated Machine Learning with Privacy-Preserving Data Processing The present disclosure provides a system and method for leakage-adaptive federated machine learning with privacy-preserving data processing. The coordination server sends to participant computing devices a global model, a training-round identifier, a challenge seed, and a privacy-policy envelope. Each device trains the model locally, divides a candidate update into parameter groups, and conducts reconstruction and inference challenges to produce a leakage-exposure profile. The profile is combined with cumulative exposure information to assign privacy zones, and compile zone-specific clipping, projection, quantisation, noise injection, encryption and secret-sharing operations. A leakage-bound verifier blocks transmission until the non-reconstructability bounds assigned to it are satisfied. Compliant updates are translated into hardware-attested capsules, and delivered as protected fragments to distributed aggregation nodes. Verified fragments are merged into independently retrievable cohort aggregates. Abnormal contributions of the cohort are detected and selectively eliminated without reconstructing the updates of individual participants. The system improves security and fault tolerance, and reduces information leakage, processing overhead and recovery time
Disclaimer: Curated by HT Syndication.
