← Home

Usable Differential Privacy: why tooling and APIs matter

Differential privacy (DP) offers provable privacy guarantees — but in practice, developers struggle to apply DP correctly. Usability gaps in libraries, unclear defaults, and sparse documentation cause subtle mistakes that can break privacy guarantees or render outputs useless.

Common practical issues

• Choosing ε (epsilon): developers need clear guidance — small values give more privacy but less utility; trade-offs must be communicated with examples.
• API ergonomics: inconsistent function names, required parameters, and missing sensible defaults make accidental misuse likely.
• Testing & auditing: DP implementations are stochastic — tooling to detect privacy regressions and to audit outputs is essential.

What helps

Documentation with practical examples, high-level frameworks (e.g., OpenDP, Google DP), and libraries that integrate with familiar ML APIs (e.g., IBM diffprivlib) lower the entry barrier.

Get involved

I’m building a community to discuss research, tooling, and reproducible evaluations. If you work on DP libraries, privacy engineering, or usable privacy, join the Discord to share ideas and collaborate: Join the Discord.

I’ll post notes from talks (NDSS USEC 2026, SOUPS), tutorials, and tooling experiments here — follow the Posts section for updates.