MUMBAI, India, Aug. 12 -- Intellectual Property India has published a patent application (202641091723 A) filed by Vaddepaty Sripradha; T. Santhosh; P. Manasa; J. Lakshmi Vydehi; and Patil Haritha on July 29, 2026, for A System And Method For Dual-Domain Conditional Gan-Based Synthetic Phishing Url And Malware Behavioral Sequence Generation For Cybersecurity Machine Learning.
Inventors include Vaddepaty Sripradha; T. Santhosh; P. Manasa; J. Lakshmi Vydehi; and Patil Haritha.
The application for the patent was published on August 07, 2026, under issue no. 32/2026.
Abstract: The present invention discloses a GPU-accelerated Dual-Domain Conditional GAN system (D2C-GAN) for simultaneously generating synthetic phishing URL feature vectors and Android malware API behavioral sequence vectors for cybersecurity machine learning training data augmentation. The invention addresses labeled data scarcity, GAN mode collapse, and cross-dataset generalization failure. The system comprises: a domain constraint extraction processor; a dual-domain conditional GAN with Shared Latent Encoder (128-dimensional), URL-Generator, Behavior-Generator, and knowledge-constrained discriminators with URL syntactic validity and API co-occurrence priors; novel evasion-diversity loss L_ED = -(D_URL_intra + D_MAL_intra) - 0.5 x (coverage_url + coverage_malware); cross-domain cosine alignment loss; three-stage quality validation pipeline; and linear normalization transfer calibration layer. Validated on PhiUSIIL (235,795 URLs) and DREBIN (15,036 APKs, 215 features): phishing detection F1 0.974, malware detection F1 0.968, cross-dataset generalization improvement 15.1%, quality validation pass rate 97.3%. Prior art search of 1,688 patents confirmed that zero patents generate both phishing URL feature vectors and malware API behavioral sequences simultaneously in a unified dual-domain framework with shared latent space.
Disclaimer: Curated by HT Syndication.
