MUMBAI, India, July 7 -- Intellectual Property India has published a patent application (202641077169 A) filed by Sr University Warangal on June 22, 2026, for Adversarial Attacks Robustness In Deep Learning Models.
Inventors include Gundapu Thirupathi; and Dr V Thirupathi.
The application for the patent was published on July 03, 2026, under issue no. 27/2026.
Abstract: The present invention is directed to a cyber-threat detection framework comprising a multi-attention deep learning architecture for detecting malicious activities under complex cybersecurity situations. This framework handles a variety of security data such as network traffic logs, system logs, endpoint telemetry, and user behavior data. The proposed hybrid learning architecture is composed of convolutional feature extraction modules and bidirectional sequence-learning networks, which are able to extract spatial and temporal features of cyber events. Multi-attention mechanism adaptively filters features relevant to security and suppresses redundant and noisy information to enhance the capability of threat discrimination. The framework includes adversarial-resilient learning strategies to make it more reliable in adversarial settings, including resistance to the presence of manipulated or perturbed inputs. The proposed system can classify threats in real time, assist in anomaly and intrusion detection and gives human-readable results using attention-based feature relevance analysis. The invention provides scalable, adaptive, and explainable cybersecurity analytics that deliver more accurate detection, lower false alarm rates, and greater resilience to more advanced attackers in enterprise, cloud, and network security infrastructures.
Disclaimer: Curated by HT Syndication.
