MUMBAI, India, June 26 -- Intellectual Property India has published a patent application (202641071321 A) filed by Rajasekhar Pillalamarri; T. Sharanya; M. Keerthana; Saranya Nair; Niji Kuriakose; Anooja B; Jisha K V; and Dr S Mohan on June 09, 2026, for Deep Learning Based System For Real Time Detection And Prevention Of Adversarial Attack In Neural Network.
Inventors include Rajasekhar Pillalamarri; T. Sharanya; M. Keerthana; Saranya Nair; Niji Kuriakose; Anooja B; Jisha K V; and Dr S Mohan.
The application for the patent was published on June 19, 2026, under issue no. 25/2026.
Abstract: The present invention relates to a deep learning-based system and method for real-time detection and prevention of adversarial attacks in neural networks. The proposed system comprises a data acquisition module, feature extraction module, adversarial detection engine, attack classification unit, dynamic defense module, model hardening unit, alert generation system, and continuous learning module. Incoming data are continuously monitored and analyzed using advanced deep learning models to identify malicious perturbations designed to mislead neural network predictions. Upon detection of a potential adversarial attack, the system automatically classifies the threat and applies suitable defense mechanisms such as input sanitization, feature squeezing, ensemble verification, confidence calibration, and adaptive retraining. The continuous learning module updates the detection framework using newly observed attack patterns, thereby improving resilience against evolving threats. The invention provides enhanced security, robustness, and reliability for artificial intelligence systems deployed in autonomous vehicles, healthcare, industrial automation, smart surveillance, IoT networks, financial systems, and other mission-critical applications requiring trustworthy neural network operation.
Disclaimer: Curated by HT Syndication.
