Privacy Accounting

Privacy accounting aims to quantify the privacy loss incurred by differentially private (DP) mechanisms, ensuring that the privacy guarantees offered by these mechanisms are accurate and reliable. Current research focuses on improving the accuracy and efficiency of privacy accounting methods, particularly for complex scenarios like adaptive compositions of mechanisms and data-dependent preprocessing steps, often employing techniques like Rényi differential privacy and privacy loss distributions. These advancements are crucial for enabling the wider adoption of DP in machine learning and other applications by providing tighter privacy bounds and reducing the overestimation of privacy loss, thereby improving the utility-privacy tradeoff.