MUMBAI, India, Sept. 7 -- Intellectual Property India has published a patent application (202641104737 A) filed by J. Hari Krishna; T. Shailaja; Morla Manasa; Nikhita Motamarri; and Siddabattula Geetha Sri Priyanka on August 31, 2026, for Adversarial Robustness Analysis In Deep Neural Networks And Advanced Defense Mechanism Enhancements.
Inventors include J. Hari Krishna; T. Shailaja; Morla Manasa; Nikhita Motamarri; and Siddabattula Geetha Sri Priyanka.
The application for the patent was published on September 04, 2026, under issue no. 36/2026.
Abstract: Deep neural networks have achieved remarkable success in many areas such as image recognition, natural language processing, and autonomous systems, yet they remain highly vulnerable to adversarial attacks. These attacks involve small, carefully designed changes to input data that are often invisible to humans but can cause the models to make completely wrong predictions. This weakness raises serious concerns about the reliability and safety of deep learning systems when used in real- world applications. This work carefully examines the adversarial robustness of deep neural networks by studying how different types of attacks, including gradient- based and black-box methods, affect model performance. It analyzes the reasons behind this vulnerability and evaluates existing defense strategies such as adversarial training, input preprocessing, regularization techniques, and certified robustness methods. Building on these insights, the study proposes enhanced defense mechanisms that combine multiple approaches to improve the models’ ability to resist a wider range of attacks while maintaining high accuracy on normal data. Through extensive experiments on standard datasets, the enhanced defenses demonstrate better resistance against both known and unseen adversarial examples compared with traditional methods. The findings highlight practical ways to make deep neural networks more secure and trustworthy, offering valuable guidance for developing reliable artificial intelligence systems in safety-critical domains.
Disclaimer: Curated by HT Syndication.
