MUMBAI, India, Oct. 12 -- Intellectual Property India has published a patent application (202641119806 A) filed by T. K. Pradeepkumar; Suguna A; B. Jaya Karuna; Arunprasad L; Muralitharan G; Dr. K. S. Kavin; Praveen Kumar B; Glastin Y V; R. Monisha; and S. Jablin Briskila on October 07, 2026, for Self-Adaptive Deep Learning System For Real - Time Malware Evolution Prediction.
Inventors include T. K. Pradeepkumar; Suguna A; B. Jaya Karuna; Arunprasad L; Muralitharan G; Dr. K. S. Kavin; Praveen Kumar B; Glastin Y V; R. Monisha; and S. Jablin Briskila.
The application for the patent was published on October 09, 2026, under issue no. 41/2026.
Abstract: The current invention introduces a self-adaptive DL system for real-time malware evolution prediction to detect, analyse, and predict malware behaviours continuously. Malware information is collected from diverse sources, including binary executables, API call sequences, network packets, process execution logs, registry modifications, file system activities, threat intelligence feeds, and sandbox behaviours. This data is then pre-processed using techniques such as feature normalization, opcode extraction, API sequence modelling, behavioural event encoding, denoising, and temporal sequence segmentation, resulting in clean multi-modal malware representations. Evolving malware behaviours are captured through dynamic behavioural sequence formation combined with self-adaptive behavioural windows. Robust malware features are generated via multi-modal feature embedding and self-supervised representation learning, enabling the system to leverage both labelled and unlabelled data sources. The MHST-DPA architecture integrates hierarchical state space learning for efficient analysis of long-sequence malware behaviours, while Transformers capture short, mid, and long-range behavioural patterns. An evolution memory bank continuously retains knowledge of malware families and their variants, while dynamic prototype adaptation updates malware family representations as new variants emerge. These enriched representations support advanced tasks such as malware family identification, variant evolution prediction, behavioural mutation prediction, attack stage prediction, and threat level estimation. Proposed system employs self-adaptive online learning to incorporate new malware samples, refine memory, adapt prototypes, and perform concept drift-aware selective re- training. To ensure transparency, explainable AI techniques are applied for interpretable malware evolution analysis, enabling security experts to identify influential behavioural patterns and relate them to previously observed malware. A threat intelligence feedback mechanism continuously updates the malware knowledge base, historical memory, and model validation metrics, ensuring sustained performance.
Disclaimer: Curated by HT Syndication.
