Differential Privacy with Fine-Grained Provenance

This technical report discusses the integration of differential privacy (DP) and fine-grained provenance to enhance privacy protection in data analysis. It outlines the challenges faced in practical implementations of DP, particularly in maintaining privacy budgets and accommodating dynamic datasets. The authors propose a taxonomy of privacy provenance, categorizing it into three types: why-DP-provenance, which clarifies the outputs to analysts; how-DP-provenance, which utilizes metadata for improved privacy; and where-DP-provenance, which monitors privacy budget consumption across varying data sources. The report reviews existing DP systems that utilize these provenance techniques and identifies open challenges and future directions for research. The aim is to foster a deeper integration of DP with data provenance frameworks to create more effective and user-friendly DP systems. The document concludes with a roadmap for further exploration in this field, emphasizing the need for optimized systems that balance privacy and utility.