1. Privacy Preserving Data
2. k-anonymous
3. Differential Privacy

(Source: https://aircloak.com/history-of-data-anonymization/)

• Massachusetts’ Group Insurance Commission", William Weld & Latanya Sweeney

• “But now that we know this, can the problem of privacy be solved by simply concealing information about birthdate, sex, and zip code in future data releases?”

• Unfortunately, although k-anonymity can prevent record reidentification in the strictest sense, it highlights that reidentification is not the only (or even the main) privacy risk.
• When multiple datasets are released, even if all of them have been released with a k-anonymity guarantee.
• Solution is not “aggregate statistics”: privacy risks there, too.

• A stringent measure of privacy but still allows insights.
• “We should be comparing what someone might learn from an analysis if any particular person’s data was included in the dataset with what someone might learn if it was not.”
  • Gödel Prize: Cynthia Dwork, Frank McSherry, Kobbi Nissim, and Adam Smith.
• Requires Randomized Algorithm: randomized algorithm maps inputs to the probabilities of different outputs
• The use of randomness - deliberately add noise to computations, in a way that promises that any one person’s data cannot be reverse-engineered from the results.

• Dwork, C. (2008, April). Differential privacy: A survey of results

• We are able to learn what we wanted without incidentally collecting any strongly incriminating information about any single individual in the population : collecting only the aggregate information that we wanted.

• an important trade-off: the more private we make the protocol, the more plausible deniability each individual we poll gets, and the less anyone can learn from an individual polling response

• So to get the same accuracy, the smaller the privacy parameter, the more people we need to poll.

• 2020 Census will be protected with differential privacy
• entralized model of differential privacy : adding privacy protections to the aggregate statistics it publicly releases
• Questions:
  • So why did the Census Bureau decide to adopt differential privacy in the first place?
    
  • Why a weaker centralized trust model, which does not protect against subpoenas, hacks, and the like?

• Fitbit & Strava

• Unintended Correlations:

• Catching The Golden State Killer:

• Q: With DP’s aggregate statistic , are the application in government use “equitable and fair”?
• Q: Can we protect privacy by not recording things that should not be recorded: race, etc. ?