MUMBAI, India, July 11 -- Intellectual Property India has published a patent application (202511060841 A) filed by Ajay Kumar Garg Engineering College, Ghaziabad, Uttar Pradesh, on June 25, for 'text attack defense system and working method thereof.'
Inventor(s) include Alok Choudhary; Harshit Singhal; Laxman Pathak; Sanyam Jain; and Dhanshri Parihar.
The application for the patent was published on July 11, under issue no. 28/2025.
According to the abstract released by the Intellectual Property India: "The present invention discloses a text attack defense system and method for safeguarding machine learning (ML) models from adversarial text manipulations. The system comprises a plurality of AI acceleration units including NPUs, FPGAs, and ASICs; a high-bandwidth memory (HBM) subsystem; and a Compute Express Link (CXL) interconnect fabric for low-latency, cache-coherent communication. Integrated secure processing units (SPUs) with Hardware Security Modules (HSMs) and Trusted Execution Environments (TEEs) ensure cryptographic integrity. The system further includes a secure IoT data ingestion layer, an AI-powered text attack detection and mitigation engine, a distributed ML orchestration framework, an AI-driven threat intelligence module, and a privacy-preserving AI/ML module employing federated learning and cryptographic methods. The method involves ingesting textual data, robust embedding, real-time adversarial detection, mitigation via denoising, automated threat response, threat intelligence updates, and privacy-preserving federated training. The invention ensures scalable, resilient, and secure ML defense across decentralized and adversarial text-rich environments."
Disclaimer: Curated by HT Syndication.
