MUMBAI, India, March 13 -- Intellectual Property India has published a patent application (202621010502 A) filed by Dr. Atulkumar Gulab Sanadi; Zaid Anjum Nisar Ahmed; Nadeem Anwer Nisar Ahmed; Mahesh Vishwambhar Ghotkar; Santosh Rameshwar Mitkari; Balasaheb Bappaji Gadekar; Dr. Nanduri Srinivas; Dr. Varsha Damodhar Jadhav; Rucha Bhausaheb Jadhav; and T. Manju, Sangli, Maharashtra, on Feb. 1, for 'mathematical foundations of neural networks and deep learning in high-dimensional data analysis.'
Inventor(s) include Dr. Atulkumar Gulab Sanadi; Zaid Anjum Nisar Ahmed; Nadeem Anwer Nisar Ahmed; Mahesh Vishwambhar Ghotkar; Santosh Rameshwar Mitkari; Balasaheb Bappaji Gadekar; Dr. Nanduri Srinivas; Dr. Varsha Damodhar Jadhav; Rucha Bhausaheb Jadhav; and T. Manju.
The application for the patent was published on March 13, under issue no. 11/2026.
According to the abstract released by the Intellectual Property India: "The present invention provides a mathematically grounded system and method for neural network based deep learning in high-dimensional data environments. The invention introduces structured neural layers that incorporate orthogonal projections, manifold-aware mappings, and norm-preserving transformations to reduce redundancy and preserve critical information. Training is performed using analytically derived optimization techniques combined with spectral norm constraints, gradient stabilization, and probabilistic regularization. Information-theoretic metrics, including entropy and mutual information, are embedded to guide feature extraction and representation compression. The framework enables end-to-end learning that improves convergence stability, generalization, interpretability, and computational efficiency across diverse high-dimensional datasets. The invention is architecture-agnostic and applicable to feedforward, convolutional, recurrent, and transformer-based networks, supporting applications in healthcare, finance, autonomous systems, natural language processing, and large-scale data analytics."
Disclaimer: Curated by HT Syndication.
