MUMBAI, India, Nov. 2 -- Intellectual Property India has published a patent application (202611110310 A) filed by Chandigarh University on September 14, 2026, for Synthetic Training Data Sanitization Using Multi-Scale Entropy And Non-Parametric Divergence Analysis.
Inventors include Shaman Sharma; and Deepak Kumar.
The application for the patent was published on October 30, 2026, under issue no. 44/2026.
Abstract: Synthetic Training Data Sanitization Using Multi-Scale Entropy and Non-Parametric Divergence Analysis relates to a computer-implemented system and method for sanitizing training data used by a machine learning model by analyzing intrinsic informational characteristics of incoming data. Multi-scale entropy characteristics are extracted at local, segment-level, document-level, semantic and temporal scales, where applicable, and compared with reference entropy characteristics associated with verified human-originated data using non-parametric divergence analysis. An authenticity score is generated and coupled to a training gate that selectively permits, attenuates or blocks model parameter updates. Borderline data may be reprocessed by a controlled variance transformation and supplied to a clean data buffer for subsequent training. A feedback monitor, audit interface and governance layer support repeated training-cycle assessment, dataset lineage and policy control, while reference characteristics can be calibrated using newly verified human data. The system is applicable to text, code, image and multimodal training data.
Disclaimer: Curated by HT Syndication.
