MUMBAI, India, Aug. 12 -- Intellectual Property India has published a patent application (202621075916 A) filed by Symbiosis International Deemed University on June 18, 2026, for Graph Neural-Network Mean-Field Reinforcement Learning System For Self-Healing Reconfigurable Intelligent Surface Control In 6g Networks.
Inventors include Dr. Akhil Gupta; Dhanish Ladwani; Salim Ansari; and Rohan Laharwani.
The application for the patent was published on August 07, 2026, under issue no. 32/2026.
Abstract: ABSTRACT GRAPH NEURAL-NETWORK MEAN-FIELD REINFORCEMENT LEARNING SYSTEM FOR SELF-HEALING RECONFIGURABLE INTELLIGENT SURFACE CONTROL IN 6G NETWORKS The present invention provides a system (100) and method for intelligent distributed control of Reconfigurable Intelligent Surfaces (RIS) in ultra-dense sixth-generation wireless networks. The system comprises a multi-antenna base station (110), distributed RIS panels (120), user equipment devices (130), a graph construction module (140) that models RIS panels as nodes in a dynamic graph, a Graph Neural Network processing module (150) that performs topology-aware message passing to generate spatial embeddings replacing conventional uniform mean-field averaging, a Mean- Field Reinforcement Learning module (160) that trains scalable distributed control policies for phase shift optimization, a self-healing module (170) that detects panel failures and dynamically reconfigures the graph and active panel operations, and a central controller (180). The system achieves thirty-five to forty percent improvement in network sum rate over conventional multi-agent approaches and eight to twelve percent over standard mean-field methods, with self-healing recovery within ten to fifteen time steps after failure. [
Disclaimer: Curated by HT Syndication.
