Differentiable and Scalable Generative Adversarial Models for Data Imputation

Data imputation has been extensively explored to solve the missing data problem. The dramatically increasing volume of incomplete makes models computationally infeasible in many real-life applications. In this paper, we propose an effective scalable system named SCIS significantly speed up training differentiable generative adversarial under accuracy-guarantees for large-scale data. consists two modules, differentiable modeling (DIM) and xmlns:xlink="http://www.w3.org/1999/xlink">sample size estimation (SSE). DIM leverages a new xmlns:xlink="http://www.w3.org/1999/xlink">masking Sinkhorn divergence function make arbitrary model differentiable, while such model, SSE can estimate appropriate sample ensure user-specified accuracy final model. Moreover, also accelerate autoencoder based models. Extensive experiments upon several datasets demonstrate that, our proposed by 6.23x. Using around 1.27% samples, yields competitive with state-of-the-art methods much shorter computation time.