Paper Source: https://arxiv.org/abs/2012.15713
Introduces Kamino, a novel system designed for differentially private (DP) data synthesis, aiming to generate artificial datasets that are both statistically useful and structurally sound. It addresses a critical flaw in existing DP synthesis methods, which often fail to preserve integrity constraints and logical relationships within structured data, rendering the synthetic output unusable.
Kamino achieves this by integrating a probabilistic database framework and a constraint-aware sampling mechanism directly into the data generation process, ensuring that synthetic data adheres to predefined rules like denial constraints (DCs).
The text highlights Kamino's superior performance in preserving data consistency and maintaining utility for machine learning tasks compared to previous methods, demonstrating that structure preservation is crucial for effective utility.
While acknowledging the computational overhead and specialized expertise required for its implementation, the document positions Kamino as a significant advancement for high-value applications in regulated industries and software testing, emphasizing the importance of ethical governance for synthetic data.
