Agenda Request - Learning using aggregation APIs

[mvono]

Agenda+: Learning using aggregation APIs

Since 2020, Criteo team has been following closely and contributing actively on reporting and learning use-cases under browser vendors APIs, including aggregation ones. One key contribution was the AdKDD 2021 privacy-preserving challenge: https://arxiv.org/pdf/2201.13123, where approaches have been proposed to learn using aggregated reports.

We would like to provide an updated perspective on hybrid (aggregated / granular data) learning taking as inputs both granular display-level events & aggregated labels (i.e., label proportions for a fixed set of features).

Time

45min including Q&A.

[AramZS]

Hi, we're a little short on time, so we're going to ask you to cut this down to 30m if that's possible to present at TPAC?

[mvono]

Hi Aram, yes, no problem.

[AramZS]

Heads up your Friday timeslot is 2:45-3:15

[csharrison]

Can we post the slides? (edit: nvm I see there is a PR #196)