MUMBAI, India, Oct. 26 -- Intellectual Property India has published a patent application (202641122964 A) filed by Bikkina Jeswanth; M Pardha Saradhi Reddy; J Veda Suhas Kulkarni; Dhiravathu Bhargav Sai; S Sanjay; Gokavarapu Sri Mani Sai Abhinav; and Jonnalagadda Surya Kiran on October 14, 2026, for Cost-Aware Adaptive Retraining Framework For Mitigating Concept Drift In Non-Stationary Streaming Systems.
Inventors include Bikkina Jeswanth; M Pardha Saradhi Reddy; J Veda Suhas Kulkarni; Dhiravathu Bhargav Sai; S Sanjay; Gokavarapu Sri Mani Sai Abhinav; and Jonnalagadda Surya Kiran.
The application for the patent was published on October 23, 2026, under issue no. 43/2026.
Abstract: COST-AWARE ADAPTIVE RETRAINING FRAMEWORK FOR MITIGATING CONCEPT DRIFT IN NON-STATIONARY STREAMING SYSTEMS ABSTRACT A cost-aware adaptive retraining framework for streaming systems mitigates concept drift by treating retraining as an economic decision. A drift monitoring module computes a drift severity score from input-distribution and prediction-quality statistics. A cost estimation module determines an expected drift cost and, for each of several candidate retraining actions of differing scope, an expected retraining cost and expected performance recovery. A decision engine selects, within a resource budget, the action with the highest net utility, and a training window selector chooses training data according to the drift type. A candidate model is validated in shadow mode against the deployed model and promoted only if a promotion criterion is met. Realized costs and gains are fed back to refine cost estimates and detection thresholds, reducing unnecessary retraining while maintaining accuracy.
Disclaimer: Curated by HT Syndication.
