Privacy by Design: Differential Privacy

Last time we saw that simple data masking often doesn't protect people as well as you'd expect. This session picks up from there with differential privacy, one of the techniques used to fix that.

The idea is that you can share useful results from a dataset without revealing whether any one person is in it. We'll go through what that means, how noise gets added to make it work, what epsilon and the privacy budget are, and how to think about the trade-off between privacy and accuracy.