The Online  Advertising industry is seeing a major shift today in its operational  constraints with a global movement towards more privacy. Popular  techniques for privacy-compliant advertising such as aggregation and differential privacy mechanisms were shown to match high privacy standards but also raise concerns  about the possibility to learn relevant machine learning models for ad  placement.

 

We propose in this challenge to explore the trade-off between privacy level and prediction performance, on data donated by Criteo -  an industry leader that already released several open datasets for research purposes. To anchor the competition in reality, the challenge  design is inspired by (and as close as possible/convenient to) current  propositions in the Privacy Sandbox discussed in the Improving Web Advertising forum at W3C.

 

The competition consists in two tasks of click and sales prediction under privacy constraints, for a total prize money of $20,000.

 

The competition website is accessible at http://go.criteo.net/criteo-privacy-preserving-ml-competition.