[HN Gopher] Understanding privacy risk with k-anonymity and l-di...
___________________________________________________________________
Understanding privacy risk with k-anonymity and l-diversity
Author : marols
Score : 65 points
Date : 2024-11-05 09:30 UTC (3 days ago)
(HTM) web link (marcusolsson.dev)
(TXT) w3m dump (marcusolsson.dev)
| p4bl0 wrote:
| People interested in this will probably also like reading this
| friendly introduction to differential privacy:
| https://desfontain.es/blog/friendly-intro-to-differential-pr...,
| which is friendly yet goes into a lot of details and techniques
| in a long series of blog posts.
| porridgeraisin wrote:
| Yes, that is a great resource. And if you're looking for a
| connection to ML, the search term is "DP-SGD".
| 3np wrote:
| Also great as well as entertaining: https://gwern.net/death-
| note-anonymity
| isaacremuant wrote:
| > Imagine you're a data analyst at a global company who's been
| asked to provide employee statistics for a survey on remote
| working and distributed teams
|
| I'm going to go on a limb and assume the exercise of the analysis
| is just for fun and that the decision was already made by the
| strategic team of people who never go to the office themselves or
| aren't really affected by it.
|
| The 0 to 3 day move that many did was not supported by data in
| any of the cases I've seen and had more to do with government
| pushes and saving downtown or encouraging self layoffs.
|
| The article itself on the technique is fine although I'd also
| make sure to include the percentage with regards to non
| participants because trust in anonymity over a contentious topic
| will definitely affect who answers and what they answer.
| FjordWarden wrote:
| No mention of ARX, but it is also a tool that lets you calculate
| those metrics: https://arx.deidentifier.org
| mentalgear wrote:
| This reminds me of when Apple first introduced, with great
| fanfare, their pivot to privacy-first and "Differential Privacy."
|
| However, when privacy experts later examined Apple's
| implementation, they found that the promised privacy was largely
| an illusion. The parameters Apple had chosen for their
| Differential Privacy were so weak that only a few data exchanges
| would be enough to de-anonymize individual users.
|
| I don't know if they improved on it, but back then it was less
| about true privacy and more about the appearance of privacy and
| an unfortunate example of marketing (core differentiator, premium
| justification) taking precedence over meaningful protection.
___________________________________________________________________
(page generated 2024-11-08 23:01 UTC)