Intellectual Property India Publishes Patent Application for Federated Artificial Intelligence System For Privacy-Preserving Collaborative Learning In Industrial Iot Networks

MUMBAI, India, Aug. 12 -- Intellectual Property India has published a patent application (202641093052 A) filed by Chethana. M; Dr. Yoganandini A P; Chethan. H S; Keerthana. A; Divya. K P; Yashodha. G; and Arunkumar. A on July 31, 2026, for Federated Artificial Intelligence System For Privacy-Preserving Collaborative Learning In Industrial Iot Networks.

Inventors include Chethana. M; Dr. Yoganandini A P; Chethan. H S; Keerthana. A; Divya. K P; Yashodha. G; and Arunkumar. A.

The application for the patent was published on August 07, 2026, under issue no. 32/2026.

Abstract: ABSTRACT OF THE INVENTION Title: A FEDERATED ARTIFICIAL-INTELLIGENCE SYSTEM AND METHOD FOR PRIVACY-PRESERVING COLLABORATIVE LEARNING IN INDUSTRIAL INTERNET-OF-THINGS NETWORKS WITH DIFFERENTIAL PRIVACY, HOMOMORPHIC ENCRYPTION, BYZANTINE-RESISTANT AGGREGATION AND AUDITABLE GLOBAL-MODEL DISTRIBUTION The invention relates to a computer- implemented federated artificial-intelligence system and method for privacy-preserving collaborative learning across a plurality of geographically-distributed and organizationally-separated Industrial Internet-of-Things participant nodes. At each participant node, a PLC/SCADA telemetry collector (101), a sensor array collector (102), an MES/quality-record collector (103) and an operator/maintenance event-log collector (104) feed an on-site data minimization and de-identification layer (105), a local artificial-intelligence model training engine (106), a differential-privacy noise injection module (107), a homomorphic encryption module (108), a secure channel and client authentication module (109) and a client capability advertiser (110). At a federated coordinator, a secure aggregation server (111) receives the encrypted gradients under a Byzantine-resistant outlier filter (112) that unconditionally quarantines any malicious or corrupted update prior to and independently of the aggregation; an adaptive client-selection orchestrator (113) accommodates heterogeneous participant capabilities across rounds; a global model integrity validator (114) verifies the aggregated model; and an immutable audit ledger (115) records every step for regulatory and inter-organizational compliance review. The framework combines on-site data minimization, differential-privacy noise injection, homomorphic encryption, secure multi-party aggregation and Byzantine resistance into a defence-in-depth privacy architecture directed at the specific ingest modalities and threat model of Industrial IoT, and is expressly a collaborative-learning framework operable under an inter-participant governance agreement and within the constraints of the applicable data-protection and industrial-cyber-security regulations. Number of Claims: 5 Number of Drawings: 3

Disclaimer: Curated by HT Syndication.