Intellectual Property India Publishes Patent Application for Unified Framework For Autonomous Concept Drift Detection, Self-Monitoring, And Adaptive Model Repair In Ai Systems

MUMBAI, India, Aug. 12 -- Intellectual Property India has published a patent application (202621075932 A) filed by Symbiosis International Deemed University on June 17, 2026, for Unified Framework For Autonomous Concept Drift Detection, Self-Monitoring, And Adaptive Model Repair In Ai Systems.

Inventors include Dr. Akhil Gupta; Janhavi Donadkar; Sanskruti Gorle; and Riddhi Thakkar.

The application for the patent was published on August 07, 2026, under issue no. 32/2026.

Abstract: ABSTRACT UNIFIED FRAMEWORK FOR AUTONOMOUS CONCEPT DRIFT DETECTION, SELF-MONITORING, AND ADAPTIVE MODEL REPAIR IN AI SYSTEMS A unified autonomous framework (100) for self-monitoring and adaptive model repair in long-lived AI systems subject to concept drift is disclosed. The framework integrates a drift detection module (110) employing multi-signal fusion of DDM, ADWIN, and Page-Hinkley detectors with repair-outcome validation to filter false alarms; a self-monitoring and ensemble coordination module (120) computing multi-scale statistics via Gauss-Markov optimal weighting with Shannon entropy and topology-aware coordination; and an autonomous repair module (130) maintaining diverse expert models updated through Regularized Diversity-Coupled Loss with Kalman-filtered variance normalization and per-expert AdaGrad Online Mirror Descent achieving superior regret bounds. A shared coordination interface (140) synchronizes all modules, escalating repair responses when simultaneous drift is detected. A two-level ADWIN trigger separates warning-level adaptation from confirmed-drift full retraining, reducing unnecessary retraining by approximately 40 percent. Prequential evaluation on climate datasets demonstrates 18.7 percent MAE reduction and 31.5 percent false alarm reduction versus baselines. [

Disclaimer: Curated by HT Syndication.