MUMBAI, India, Feb. 13 -- Intellectual Property India has published a patent application (202611000383 A) filed by Delhi Technological University, New Delhi, on Jan. 2, for 'system for detecting and mitigating poisoning attacks in an artificial intelligence model.'
Inventor(s) include Kumar, Rajeev; Kumar, Vinod; Yadav, Rajesh Kumar; Beniwal, Rohit; Kumar, Rahul; and Sharma, Deepak.
The application for the patent was published on Feb. 13, under issue no. 07/2026.
According to the abstract released by the Intellectual Property India: "Embodiments of the present disclosure relate to a system (100) and a method (200) of detecting and mitigating poisoning attacks in an artificial intelligence model. The system (100) coordinates operations by receiving input samples through an input interface (104). A spatial perturbation unit (106), a frequency perturbation unit (108), and a semantic perturbation unit (110) generate domain-specific perturbed samples. A model inference unit (112) processes clean and perturbed samples to produce corresponding outputs. A sensitivity computation unit (114) derives domain-wise sensitivity values representing output deviations. A correlation analysis unit (116) evaluates pairwise dependencies among sensitivity values across domains. An entanglement matrix generation unit (118) assembles dependency values into a structured matrix. An eigenvalue analysis unit (120) extracts eigenvalues representing cross-domain coupling characteristics. The system (100) aggregates eigenvalue deviations into a scalar entanglement index and initiates corrective actions when abnormal dependency is identified."
Disclaimer: Curated by HT Syndication.
