New & scary paper by Ryan Steed, Diana Qing, and Steven Wu, that presents a novel application of reconstruction attacks on Census data: identifying households that live in violation of occupancy guidelines 😨 This matches one of the fears that some census participants have: what if their landlord, a custodial parent, or a government agency uses census data against them, in an eviction case? Preventing this from happening is critical to protect respondent's privacy, and ensure that people feel safe enough to respond truthfully in future census surveys 💡 The conclusion of the research is unsurprising: the attack has a very high success rate when using legacy disclosure avoidance methods, but is well-mitigated when using differential privacy ✅ Full paper ➡️ https://lnkd.in/eyn6wxzT 📜
The F1 value for the "No protections" case is 68.4 and DP is 67.2. While the significantly lower precision, closer to 50% for the DP looks good on paper, in practice, that increased Recall is a bit scary, since a govt. agency can still choose to audit all the households, and catch roughly the same number of households violating the policy. In general, over-focus on precision as a metric in the privacy literature worries me. Yes, there's plausible deniability with low precision, but it doesn't completely measure risk. I still favor using DP, just pointing out how choice of metrics is important here.
There’s plenty of attacks. Just so everyone has a notion: Most HR software already has access to these databases. We then combine household value, revenue, to people using public available data. Then we apply some algos to extract how much is this person willing to accept as a paycheck. This is then sold as a solution. When we say “surveillance economy”, privacy and data advocates are not making this up. There’s way too much data, way too much abuse and way too much Wild West abuse; because companies know they can get away with it. And it compensates financially to abuse single humans at scale. Problem is, HR is but a dot; in the whole ecosystem of surveillance and corporate use of data in unethical ways. All Fortune 500 are doing it.