Hacker News Reader

Interesting, but what's to prevent the person selling the data from completely fabricating it (as opposed to following noise according to some differential privacy algorithm)? Or is the assumption that enough data owners are truthful that you could filter out this behaviour?

View on HN

Could this be a useful model to apply to analytics and tracking?

Selective sampling is already common (run X diagnostic 10% of the time). Could you say to a user of your application "Hey, we are tracking these X parameters, but applying Y degrees of uncertainty to your individual data."

I'm not pro-tracking by any means, but would this be a reasonable improvement to users' privacy without first needing to convince companies to stop all tracking wholesale?

Reply on news.ycombinator.com