MUMBAI, India, May 29 -- Intellectual Property India has published a patent application (202641061209 A) filed by Vellore Institute Of Technology, Vellore, Tamil Nadu, on May 14, for 'a self-healing neural network system for detecting model drift and performing structural recovery in a real-time artificial intelligence environment.'
Inventor(s) include Vedansh Gupta; Kushagra Mukhija; and Dr. Balakrushna Tripathy.
The application for the patent was published on May 29, under issue no. 22/2026.
According to the abstract released by the Intellectual Property India: "The present invention relates to a self-healing neural network system and method for real-time detection of model drift and autonomous recovery in artificial intelligence environments are disclosed. The invention employs continuous monitoring of runtime performance indicators, including prediction accuracy, confidence distribution, hidden-layer activations, and error rates, to identify deviations from baseline behavior using a drift detection engine. Upon detection of drift or collapse tendencies, a diagnosis engine determines underlying causes such as neuron inactivity, overactivity, feature irrelevance, noise interference, and structural bottlenecks. Based on the diagnosis, an autonomous repair controller initiates corrective actions through structural modifications, including neuron addition, pruning, layer expansion, and activation-pathway rerouting, or parameter-level adjustments such as selective retraining and weight reinitialization. The repaired model is validated against predefined performance thresholds using live or shadow data, ensuring only effective configurations are retained, while unsuccessful repairs trigger reversion to a stable prior state. A knowledge repository stores historical drift patterns and repair outcomes to enhance future adaptation, thereby enabling continuous, resilient, and automated optimization of neural networks without manual intervention or full retraining."
Disclaimer: Curated by HT Syndication.
