Formal Methods and Proofs of Privacy Properties, Part I
summary
Three-hour mini-course that claims formal methods can make differentially private algorithms correct by design and by proof, with concrete code and tools. It shows a CDF example implemented two ways (sequential Laplace noise vs partitioned counts), explains PinQ and iRavat style systems, and demos a privacy budget agent that blocks queries when exhausted and a noisyCount primitive that adds Laplace noise. Covers concrete failures (sparse vector variants, floating-point leaks) and the real challenges: relational, probabilistic and quantitative reasoning for end-to-end verification.
AI-written summary · about
chapters (11)
- 0:00Start
- 1:25Introduction to Formal Methods
- 3:47High-Level Introduction of What Formal Methods Are
- 4:58The Goal of Formal Method
- 5:08Why Correctness Matters
- 14:08Preconditions and Postconditions
- 29:52Example of a Sparse Vector
- 34:54What Are the Challenges of Verification for for Verification of Differential Privacy
- 36:51Approach Is Designing a Language Which Guarantees Differential Privacy by Design
- 44:59Build a private version of cdf
- 46:36Parallel Composition
related talks
29mPrivacy-preserving methods: Building secure projects
26mAaron Roth, Associate Professor, University of Pennsylvania, NYC 2017
45mMaking Privacy Technology Accessible: Benchmarks and Platforms
01h 16mLaw and CS
22mPrivacy-preserving Computation in Plain Sight
29mVerifying Pufferfish Privacy in Hidden Markov Models
01h 00mPANEL: Statistical Theory, Privacy and Data Analysis
42mPrivacy Challenges in Machine Learning