Privacy Workshop Summary

Two weeks ago, on April 24, thirty-some people gathered at Caltech for a workshop on privacy. At least from my (biased) perspective, it was a great time!

The basic format was that each speaker was paired with a discussant who doesn’t work on privacy, but who could provide motivation, discussion, context, or critique for the talk.

Moritz Hardt gave a talk on privacy + adaptive data analysis, and he shared the session with Antonio Rangel, a Caltech neuroeconomist. They held an exciting and informative conversation on the the risks of false discovery in data analysis, and the potential of new, differential privacy-based tools to mitigate those risks.

Kobbi Nissim gave a talk on privacy + learning, and he shared the session with Pietro Perona, a Caltech electrical engineer who studies vision. One interesting thing that has come out of recent work on private learning is that the central challenge for many tasks is to privately learn the underlying “scale’’ of the data, and there has been interesting progress on private scale-finding in the past few years.

Aaron Roth spoke on privacy as a toolkit for mechanism design in large games. He shared the session with Leeat Yariv, a Caltech economist. Together, they helped us understand the potential of the privacy toolkit to provide robustness guarantees that could be appealing in many settings where individuals don’t have huge influence on others (e.g., traffic, large matchings).
The program is available here:

and there are links with slides and video!

On Saturday morning after the workshop, we went for a “privacy hike” on nearby Verdugo Mountain. The signs promised mountain lions, but the lions decided to keep to themselves.
Privacy HikeThanks, everyone, for a fun workshop!