Composable and versatile privacy via truncated CDP

We propose truncated concentrated differential privacy (tCDP), a refinement of differential privacy and of concentrated differential privacy. This new definition provides robust and efficient composition guarantees, supports powerful algorithmic techniques such as privacy amplification via sub-sampling, and enables more accurate statistical analyses. In particular, we show a central task for which the new definition enables exponential accuracy improvement.

Authors

• Mark Bun
• Cynthia Dwork
• Guy N. Rothblum
• Thomas Steinke 0002

Keywords

• algorithmic stability
• differential privacy
• subsampling