ML challenge on aggregate measurement

Hi everyone,

I wanted to publicize a challenge that we at Criteo are launching at adKDD (see info on adKDD https://www.adkdd.org/).

This challenge aims at solving the optimisation use case (learning models) using aggregate reporting with differential privacy, which is the one that is proposed in the Privacy Sandbox.

I opened an issue giving a little more detail here<https://github.com/WICG/turtledove/issues/171>. If you are interested in these topics, we strongly encourage you to participate!

We will release a dataset that will be used during the challenge. The challenge is open to every data scientist, inside and outside adtech. As per usual during these competitions, there will be some money prize for the winners :) .

We hope this challenge will help everyone get a better understanding on some of the choices that have been discussed in this forum, in an open and transparent way. We also hope that ultimately it could inform some of the remaining choices that are to be made on the design, e.g on the noise level.

If there is some time left in the agenda, I will be delighted to speak about it at the next meeting.

Best,
Basile Leparmentier