AMD-GAN: Class-Specific Adaptive GANs for Configurable Synthetic Traffic and Imbalance Correction

AMD-GAN: Adaptive Manifold-Decoupled GAN for NIDS. Solves extreme class imbalance via isolated generators and dynamic training physics (size-dependent regularization). Validated on CIC-IDS2017. A project part of a PhD. financed by INCIBE and CiberIA Project. Estos resultados han sido (parcialmente) financiados por la Cátedra Internacional UMA 2023, la cual forma parte del Programa Global de Innovación en Seguridad para la promoción de Cátedras de Ciberseguridad en España financiado por la Unión Europea Fondos NextGeneration-EU, a través del Instituto Nacional de Ciberseguridad (INCIBE). These results have been (partially) funded by the UMA 2023 International Chair, which forms part of the Global Security Innovation Programme for the promotion of Cybersecurity Chairs in Spain, funded by the European Union’s NextGeneration-EU funds, through the National Cybersecurity Institute (INCIBE).