Differential Privacy, the Laplace Mechanism and Privacy-Budget Accounting: Publishing Business Statistics without Exposing Any Individual: chapter answers
When owners publish figures they usually follow three unwritten rules: deleting names and phone numbers makes data anonymous, publishing only aggregates exposes no one, and a little jitter sized by feel settles any remaining doubt. This book refutes all three and offers, through differential privacy, an alternative that can be recomputed, audited and overturned. Sweeney estimated that about 87 % of Americans are unique on ZIP code, sex and date of birth, and Golle's re-estimate is about 63 %; the book's cell model shows that in a postcode of ten thousand about 0.843 of residents are unique on…
DOI pending · monograph page goes live after publication · 5DL-B-296 · 中文版
- Chapter 1 · "Anonymous" Is Not Anonymous: Quasi-Identifiers, Differencing and Reconstruction
Deleting names does not make data anonymous: in a postcode of ten thousand, about 0.843 of residents are unique on date of birth and sex, and only past forty thousand does that fall to one half; Sweeney and Golle report… - Chapter 2 · The Laplace Mechanism: Epsilon Is a Ceiling on the Likelihood Ratio
Chapter 2 defines the Laplace mechanism and epsilon: a count at epsilon 1 has a 95 % half-width of about 2.996 people, and no adversary can beat guessing by more than 0.4621. - Chapter 3 · Sensitivity: Clip Revenue First, and Spend the Budget Once on a Customer Profile
Chapter 3 deals with sensitivity: revenue clipped per person has a root mean square error of 6545 yuan in the illustrative data, 4.09 % of revenue, and a customer profile spends the budget only once by parallel composit… - Chapter 4 · Composition and Budget Accounting: Basic, Advanced and Optimal Books
Chapter 4 computes basic, advanced and optimal composition side by side; for a daily release over a year the optimal bound is only 0.283 of the basic one. - Chapter 5 · Randomized Response and the Local Model: Letting Customers Add the Noise
Chapter 5 covers randomized response and the local model, whose error at the same epsilon is 24.6 times that of the central model. - Chapter 6 · How Large Should Epsilon Be: the Dispute over the 2020 United States Census and the Price of Small Numbers
Chapter 6 reads the choice of epsilon through a posterior ceiling and a table of relative errors, and sets out the dispute over the total budget of 19.61 for the redistricting data of the 2020 United States census: at e… - Chapter 7 · Classical Sources East and West: Han Feizi on the Difficulty of Persuasion and Warner's Randomized Response
Chapter 7 examines "words that touch the hidden matter" in the Han Feizi and Warner's randomized response of 1965 (a spinner probability of 0.8 implies an epsilon of 1.386). - Chapter 8 · A Noisy Release Procedure for the Public Statistics of a Pay-per-Use Service, and a Plan for Refutation
Chapter 8 gives a seven-step noisy release procedure for the public statistics of a pay-per-use service, the distilled tool, the Privacy Budget Accountant, an audit method (100 trials catch an implementation with halved…
Illustrative model calculation, not a live quote