Probabilistic Coupled Generative Adversarial Networks for Terahertz Antenna Multi‐Parameter Co‐Design

This paper proposes a probabilistic coupled generative adversarial network (PCGAN) integrating Gaussian process regression (GPR) with adversarial learning for terahertz antenna co‐design under limited datasets. The framework employs SMOTE‐expanded geometric parameters and GPR‐ANN coupled predictions to generate high‐fidelity datasets. A probabilistic discriminator authenticates data while learning performance‐to‐geometry mappings through iterative refinement, enabling precision geometry prediction from specifications. Experimental validation reveals 1.23% mean squared logarithmic error in geometric predictions. PCGAN completes optimization in 7.1 h, providing a solution that resolves efficiency‐accuracy conflicts in terahertz antenna design.