Deep Generative Sampling in the Dual Divergence Space: A Data-efficient & Interpretative Approach for Generative AI