Designing an Open-Source Platform for Differentially Private Analytics That Is Usable...
summary
Tumult Analytics is an open‑source platform for configurable central differential privacy used by IRS, US Census Bureau, and Wikimedia that claims provable guarantees while remaining usable for non‑experts. It exposes a dataframe‑like Python API with session privacy budgeting (example: median by ZIP code) and a composable Tumult Core that declares inspectable privacy properties and supports new mechanisms and composition, including fixes for floating‑point vulnerabilities. Integrated with Apache Spark, it scales linearly to ~1 billion rows and can produce 100s of millions of statistics.
AI-written summary · about
chapters (6)
related talks
47mOpen DP: A Proposal for an Open-Source Suite of Differential Privacy Tools
14mChaos Genius: Open Sourcing Outlier Detection and Root Cause Analysis
29mWhat We’re Doing in Australia – An API for Confidentialized Contingency Tables
44mTournesol An Open Source Project for Collaborative Algorithmic Governance
24mBuilding Open Data Platforms, an Insider's Perspective
41mOpen Source Vermont Data Platform: Access, Analysis, and Visualization
07mDriving open source adoption within the enterprise
24mState of the Art Open Source Software for AI Fairnessevaluation