MUMBAI, India, Sept. 14 -- Intellectual Property India has published a patent application (202611087326 A) filed by Dr. Sandeep Rajani on July 16, 2026, for A System And A Method For Adaptive Neural Network Training And Multi-Dimensional Model Evaluation.
Inventor includes Dr. Sandeep Rajani.
The application for the patent was published on September 11, 2026, under issue no. 37/2026.
Abstract: The present disclosure relates to a system and method for adaptive neural network training and multi-dimensional model evaluation. The system comprises a processor configured to receive training datasets, generate a dataset reliability profile, and dynamically organize training samples using adaptive curriculum learning based on prediction confidence and learning complexity. The processor further performs neural architecture evolution, gradient stability monitoring, confidence-guided sample weighting, adaptive hyperparameter optimization, and multi-objective loss optimization to improve training efficiency and model convergence. The system further preserves previously acquired knowledge during continual learning, verifies explainability consistency, predicts data distribution shifts, optimizes hardware resource utilization, and authenticates model integrity using cryptographic verification. The trained model is evaluated using multiple deployment-oriented metrics including robustness, uncertainty calibration, fairness, computational efficiency, and deployment readiness to generate a Model Readiness Index. Based on the generated index, the system autonomously determines retraining requirements and prepares an optimized neural network model for secure and reliable deployment across diverse artificial intelligence applications.
Disclaimer: Curated by HT Syndication.
