MUMBAI, India, Aug. 17 -- Intellectual Property India has published a patent application (202641095555 A) filed by Institute Of Engineering And Management, School Of University Of Engineering And Management on August 07, 2026, for Neuro Symbolic Unlearning.
Inventors include Nilanjan Chatterjee; Bapi Biswas; Sudipto Dhar; Buddhadeb Pradhan; Arijit Ganguly; Sumit Anand; Subhojit Paul; Anay Ghosh; and Niladri Sekhar Paul.
The application for the patent was published on August 14, 2026, under issue no. 33/2026.
Abstract: The present invention relates to artificial intelligence, machine learning, and trustworthy AI, particularly to a neuro-symbolic unlearning framework capable of selectively removing learned information from trained neural network models while preserving overall predictive performance and symbolic reasoning capabilities. The proposed system integrates deep neural networks with symbolic knowledge graphs, rule engines, causal reasoning modules, and explainable decision layers to identify, isolate, and erase specific knowledge corresponding to targeted data, individuals, concepts, or regulations without requiring complete model retraining. The invention employs semantic dependency analysis, symbolic trace generation, gradient localization, knowledge influence mapping, adaptive parameter correction, and post- unlearning verification mechanisms. The framework ensures regulatory compliance with privacy requirements, minimizes catastrophic forgetting, maintains reasoning consistency, and supports continual learning. The invention is applicable in healthcare, finance, legal systems, autonomous vehicles, defense, cloud intelligence, and privacy-preserving AI applications where selective forgetting is mandatory.
Disclaimer: Curated by HT Syndication.
