MUMBAI, India, Sept. 22 -- Intellectual Property India has published a patent application (202641110298 A) filed by Vellore Institute Of Technology on September 14, 2026, for System And Method For Rapid Machine Unlearning In Digital Twins For Continuous Fleet Governance And Adversarial Verification.
Inventors include Dr M Rajasekhara Babu; and Dhanush Konduru.
The application for the patent was published on September 18, 2026, under issue no. 38/2026.
Abstract: The present invention provides an AeroForge system (100) for continuous governance and selective machine unlearning of prognostic models used for aerospace propulsion fleets. A knowledge manager (130) associates ingested telemetry with a contributing-party identifier, sensitivity classification, and unlearning group identifier and stores the telemetry in an indexed knowledge repository (140). In response to an erasure request (175), a selective unlearning engine (170) identifies an exact telemetry set and modifies a predictive model (150) without complete retraining. A verification engine (180) evaluates the resulting candidate model using forgetting, utility, trust, adversarial recovery, and membership-indistinguishability measurements, including comparison with a reference model retrained without the identified party. An acceptance and deployment gate (190) selects sensitivity-dependent thresholds and prevents the candidate model from entering an operational inference path until verification requirements are satisfied. An attribution engine (160), operator console (800), and compliance record generator (195) provide prediction explainability, fleet visualization, and digitally signed erasure records, respectively, while maintaining uninterrupted fleet monitoring
Disclaimer: Curated by HT Syndication.
