3 ms·

It seems as thought this author has never read any materials on inference control... Ross Anderson's Security Engineering Chapter 9 is all about this https://w

It seems as thought this author has never read any materials on inference control... Ross Anderson's Security Engineering Chapter 9 is all about this https://www.cl.cam.ac.uk/~rja14/Papers/SEv2-c09.pdf https://www.cl.cam.ac.uk/~rja14/Papers/SEv2-c09.pdf (Anderson releases the textbook free online) This is a fairly well developed field (i.e., in medicine there highly screened "Medical Statisticians" who give data to research groups only after it's properly anonymized. They know beforehand what the researchers are interested in, and introduce noise that effects all the measures _except_ the ones the researchers said they were interested in looking at beforehand. > The idea behind the differential privacy strategy, as I understand it, is that you can basically introduce statistical noise into the block-level results in a way that will make it impossible to reverse-engineer the demographic profile of people who live at specific addresses. >This has, I think, roughly zero real-world privacy value This was painful to read...

bpodgursky 5y ago
There is an enormous practical difference between allowing inference of medical conditions (PHI deidentification) and allowing in edge cases the inference of information that is already available in public voter registration rolls.
- giaour 5y ago
  > allowing in edge cases the inference of information that is already available in public voter registration rolls. This is a really inaccurate way to characterize what researchers were able to obtain from reidentification attacks on block-level data from the 2010 census: https://www.science.org/content/article/can-set-equations-keep-us-census-data-private https://www.science.org/content/article/can-set-equations-ke...
droidist2 5y ago
> They know beforehand what the researchers are interested in, and introduce noise that effects all the measures _except_ the ones the researchers said they were interested in looking at beforehand. Isn't there a danger in the census statisticians doing the same thing? That they'll introduce noise that affects looking at the data in ways that they didn't anticipate or allow for. Or even that hinders drawing conclusions that go against current dogma.

bpodgursky 5y ago

There is an enormous practical difference between allowing inference of medical conditions (PHI deidentification) and allowing in edge cases the inference of information that is already available in public voter registration rolls.

giaour 5y ago
> allowing in edge cases the inference of information that is already available in public voter registration rolls. This is a really inaccurate way to characterize what researchers were able to obtain from reidentification attacks on block-level data from the 2010 census: https://www.science.org/content/article/can-set-equations-keep-us-census-data-private https://www.science.org/content/article/can-set-equations-ke...

droidist2 5y ago

> They know beforehand what the researchers are interested in, and introduce noise that effects all the measures _except_ the ones the researchers said they were interested in looking at beforehand. Isn't there a danger in the census statisticians doing the same thing? That they'll introduce noise that affects looking at the data in ways that they didn't anticipate or allow for. Or even that hinders drawing conclusions that go against current dogma.