MUMBAI, India, Jan. 23 -- Intellectual Property India has published a patent application (202511122311 A) filed by K. R. Mangalam University, Gurugram, Haryana, on Dec. 5, 2025, for 'system for neural architecture optimization using reinforcement feedback.'
Inventor(s) include Dr. Pankaj Agarwal; and Dr. Pawan Kumar.
The application for the patent was published on Jan. 23, under issue no. 04/2026.
According to the abstract released by the Intellectual Property India: "The present invention provides a system for neural architecture optimization using reinforcement feedback, designed to enable neural networks to autonomously evolve their structural configuration based on performance outcomes, contextual conditions, and multi-objective reward signals. The system incorporates a reinforcement-driven optimization controller, a neural architecture representation module, an adaptive transformation engine, a contextual sensing subsystem, and a performance evaluation and reward mechanism. The controller processes encoded architectural states, environmental data, and evaluation metrics to determine structural modifications that are applied through the transformation engine. The architecture undergoes iterative adaptation enabled by continuous reward feedback, thereby supporting dynamic refinement during training and live deployment. The system facilitates resource-aware optimization across heterogeneous hardware environments and supports distributed and federated learning setups. Through reinforcement-guided evolution, the invention creates an adaptive, self-improving neural architecture capable of sustaining long-term performance, efficiency, and versatility across diverse computational domains."
Disclaimer: Curated by HT Syndication.
