Data Privacy: Foundations and Applications Boot Camp
2019-01-28..2019-02-01
Berkeley, United States
who it's for
privacy researchers, applied statisticians, machine learning engineers, data scientists, federal statistical agency staff, data stewards and curators, product designers and engineers
topics
differential privacy composition accounting; Gaussian mechanism for queries; private principal component analysis; private empirical risk minimization and noisy SGD; privacy moments accountant and privacy amplification; local differential privacy and randomized response; differentially private synthetic data methods; private hypothesis testing and confidence intervals; reconstruction attacks and k-anonymity failures; matrix mechanism and workload strategies; noise-projection and semidefinite relaxations; formal verification of privacy properties
| title | author | length |
|---|---|---|
| Perspective | Kobbi Nissim | 01h 43m |
Machine attacks reconstruct records from published statistics—experiments report ~96–99% recovery and a simple birthday-style predicate singles out a random… | ||
| Large-Scale Private Learning, Part I | Kunal Talwar | 01h 02m |
Covers concrete mechanisms and bounds: the Gaussian mechanism adds N(0,σ^2) with σ = Θ(√(ln(1/δ))/ε) to any ℓ2‑sensitive query, privacy tracked via the… | ||
| Large-Scale Private Learning, Part II | Kunal Talwar | 01h 07m |
Noisy SGD uses per-example gradient clipping and Gaussian noise with privacy amplification by sampling to get utility near nonprivate models; theoretical… | ||
| Large-Scale Private Learning, Part III | Ilya Mironov | 58m |
Defines and contrasts local differential privacy vs central: accuracy degrades — sum estimation error grows roughly like sqrt(n)/epsilon rather than depending… | ||
| Perspective: Achieving Meaningful Privacy in Technology and Design | Helen Nissenbaum | 01h 14m |
Argues privacy is about privacy = appropriate information flow, not secrecy, and formalizes that with five parameters—subject, sender, recipient, information… | ||
| Statistical Inference and Privacy, Part I | Sesa Slavkovic | 01h 07m |
Differential privacy adds explicit noise (Laplace around a truncated mean) so standard estimators give biased, anti-conservative p‑values and confidence… | ||
| Statistical Inference and Privacy, Part II | Gautam Kamath | 01h 03m |
Differential privacy is applied to classical statistical tasks with concrete costs and fixes: identity testing needs Θ(√K/γ^2) samples nonprivately but privacy… | ||
| Statistical Inference and Privacy, Part III | Rachel Cummings, Sesa Slavkovic | 01h 14m |
Three concrete recipes for differentially private synthetic data: smallDB enumerates all databases of size ~O(log|F|/α^2) and samples via the exponential… | ||
| The Social Survey Statistician's Perspective | Frauke Kreuter | 01h 10m |
Surveys used by federal statisticians show concrete limits for DP: randomized response fails at scale because respondents misunderstand and default to denying… | ||
| Simons Institute Live Stream | Gerome Miklau, Sasho Nikolov | 01h 03m |
Covers concrete algorithms and bounds for answering counting (linear) queries under differential privacy: the Gaussian noise mechanism adds iid Gaussian scaled… | ||
| Algorithms for Answering Linear Queries, Part II | Gerome Miklau, Sasho Nikolov | 01h 11m |
Adds Gaussian noise to query answers and then projects back onto the feasible set of answers so noisy outputs become consistent; this Gaussian noise +… | ||
| Algorithms for Answering Linear Queries, Part III | Gerome Miklau, Sasho Nikolov, Ryan McKenna | 01h 01m |
Presents a scalable variant of the matrix mechanism that replaces a global semidefinite program with local gradient optimization over a parameterized family of… | ||
| Formal Methods and Proofs of Privacy Properties, Part I | Marco Gaboardi | 58m |
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… | ||
| Formal Methods and Proofs of Privacy Properties, Part II | Marco Gaboardi | 56m |
Language called PINQ implements privacy by creating chained agents that track budget and apply Laplace noise to counts after checking budget and throwing an… | ||
| Formal Methods and Proofs of Privacy Properties, Part III | Marco Gaboardi | 01h 02m |
Claims a practical verification path for differential privacy by encoding relational proofs into SMT solvers and using approximate probabilistic coupling rules… | ||
| Markets and Incentives for Transacting on (Personal) Information, Part II | Katrina Ligett | 01h 00m |
Walks through concrete economic models applied to personal data: search-cost models show centralization reduces transaction costs; Akerlof’s market for lemons… | ||
| Markets and Incentives for Transacting on (Personal) Information, Part III | Katrina Ligett | 01h 05m |
Differential privacy provides a formal guarantee that an epsilon change barely changes individual expected utility, so mechanisms satisfying… | ||
| Markets and Incentives for Transacting on (Personal) Information, Part I | Katrina Ligett | 58m |
Proposes a practical model where a data co-op aggregates members’ disparate personal records, negotiates access with buyers, and sells aggregate products while… | ||
| nothing matches | ||