MUMBAI, India, Oct. 26 -- Intellectual Property India has published a patent application (202611113012 A) filed by Chandigarh University on September 21, 2026, for Adaptive System For Bias-Driven And Utility-Aware Controlled Memory Decay In Machine Learning Models.
Inventors include Hitesh Kumar; Sachin Yadav; and Dr. Prabhjot Kaur.
The application for the patent was published on October 23, 2026, under issue no. 43/2026.
Abstract: Adaptive System for Bias-Driven and Utility-Aware Controlled Memory Decay in Machine Learning Models relates to machine learning model lifecycle management. A deployed machine learning model is monitored during inference to identify learned parameter paths and to obtain Usage, Bias Risk and Staleness scores. A Memory Decay Engine combines the scores in a configurable Memory Decay Index and compares the index with decay thresholds. When a decay condition is satisfied, a Model Interaction and Attenuation Module identifies parameters associated with the learned pattern and selectively reduces their values without complete model retraining. A Compliance and Audit Module records forgetting events, while subsequent monitoring provides a feedback loop for continued model maintenance. The system is implemented through computing infrastructure supporting inference, telemetry aggregation, model-parameter storage and selective parameter modification.
Disclaimer: Curated by HT Syndication.
