Adaptive Synthetic Population Generation Using Gibbs Sampler

This technical report presents an adaptive approach to synthetic population generation utilizing a one-step Gibbs Sampler. The primary objective is to maintain synthetic data relevance by integrating new information adaptively, rather than requiring complete regeneration with each update. The report compares traditional independent regeneration methods with the proposed adaptive generation, demonstrating that the latter achieves similar accuracy levels more efficiently. It highlights the effectiveness of the adaptive generator, particularly when initial data is scarce or biased, as it enriches the population sample adaptively. The authors introduce a Gibbs-resampling technique to correct errors and enhance the representativeness of synthetic data. The report also discusses the limitations of existing synthetic generation methods, which often produce static snapshots that quickly become outdated. By employing Bayesian methods, the report outlines how the adaptive approach can evolve synthetic populations in response to real-world changes, ensuring continued applicability for long-term projections.