Privacy on Beam - E2E Differential Privacy Solution for Apache Beam
summary
Privacy on Beam is an end-to-end differential privacy library for Apache Beam (currently Go, Python in progress) that implements calibrated noise, contribution bounding (e.g. max partitions contributed and per-partition clamping), partition-selection math, and mitigations for floating-point sampling attacks. It preserves utility on large pipelines—demo shows noisy visits-per-hour histograms that keep lunch/dinner peaks—ships with unit, end-to-end and stochastic tests, and targets seamless Beam-style APIs so pipelines work at scale.
AI-written summary · about
chapters (9)
- 0:00Introduction and context
- 1:54Challenges of anonymization
- 6:33Understanding differential privacy
- 10:18Privacy on beam framework
- 13:21Technical implementation hurdles
- 17:41Development principles and testing
- 21:23Codelab demonstration
- 28:01Future developments and community
- 31:26Q&A and concluding remarks
related talks
45mMaking Privacy Technology Accessible: Benchmarks and Platforms
20mShare data without revealing personal information
19mPractical Considerations for Differential Privacy
27mPrivacy-preserving statistical analyses
14mToward Provably Private Insights into AI Use
20mDesigning for User Privacy: Integrating Differential Privacy into Ad Measurement...
27mYou've got trust issues, we've got solutions: Differential Privacy
30mAccuracy First: Selecting a Differential Privacy Level for Accuracy-Constrained...