MUMBAI, India, Aug. 24 -- Intellectual Property India has published a patent application (202641097300 A) filed by Iiitb Comet Foundation; and Indian Institute Of Technology Roorkee on August 12, 2026, for Self-Supervised Fitness-Driven Method For Training Generative Models.
Inventors include Kalpeshkumar Patel; and Dr. Ekant Sharma.
The application for the patent was published on August 21, 2026, under issue no. 34/2026.
Abstract: ABSTRACT SELF-SUPERVISED FITNESS-DRIVEN METHOD FOR TRAINING GENERATIVE MODELS A method for wireless network optimization addresses non-convex optimization problems without relying on external datasets. The method comprises receiving conditional information characterizing an optimization problem for a wireless communication system, generating samples using a diffusion model, and computing fitness values using an optimization fitness function. Weights are assigned to the samples based on their respective fitness values, wherein samples having higher fitness values are assigned correspondingly greater weights. Top-k samples having highest fitness values are identified, and normalized weights are computed for the top-k samples. Model parameters are updated using the normalized weights and a loss function. The process is iteratively repeated until convergence, wherein the diffusion model is trained using data generated by the diffusion model itself. The method is applicable to reconfigurable intelligent surface phase-shift optimization in cell-free massive multiple-input multiple-output systems (Fig. 1).
Disclaimer: Curated by HT Syndication.
