Created: February 14, 2021
Modified: February 14, 2021
Modified: February 14, 2021
privacy
This page is from my personal notes, and has not been specifically reviewed for public consumption. It might be incomplete, wrong, outdated, or stupid. Caveat lector.- It seems like there is, or can be, a virtuous relationship between privacy and generalization. You don't want to memorize too many irrelevant details of a particular case. generalization is about knowing the things that will apply to a new case, a new person.
- In practice, you may have data coming in over time, and you want to generalize the trajectories. So you need to maintain a notion of identity to associate data into a trajectory, before you blur your eyes to see only what is generalizable about the trajectory.
- Unfortunately you will in general not know which features of someone will generalize until you've seen a lot of data. So privacy can be a property of the outcome of batch learning, but in online learning you may need to remember every detail for a while until you can form appropriate abstractions.