Enhancing Diversity and Feasibility: Joint Population Synthesis from Multi-source Data Using Generative Models

Researchers Farbod Abbasi, Zachary Patterson, and Bilal Farooq developed a novel WGAN (Wasserstein Generative Adversarial Network) with a custom regularization term for joint population synthesis. Their method integrates multi-source datasets simultaneously, improving recall by 10% and precision by 1% over sequential baselines. This generates more diverse and feasible synthetic populations, which are critical, realistic inputs for agent-based models (ABMs) used in transportation and urban planning simulations.