W3C home > Mailing lists > Public > public-web-adv@w3.org > October 2020

New Proposal: Evaluation of Cohort assignment algorithms for the FLoC API

From: Deepak Ravichandran <deepakr@google.com>
Date: Wed, 21 Oct 2020 19:08:11 -0700
Message-ID: <CAHtKQycNP2qMwVuq7=cbVuS1iWY7_xy5v0O+nVk9pjrP88gHTA@mail.gmail.com>
To: public-web-adv@w3.org
Hello W3C Web Advertising Group,

Google Research & Ads team added a new proposal to the github repository
regarding evaluation of Cohort assignment algorithms for the FLoC API.

[Coincise version
<https://github.com/google/ads-privacy/blob/master/proposals/FLoC/README.md>
, Detailed White paper
<https://github.com/google/ads-privacy/raw/master/proposals/FLoC/FLOC-Whitepaper-Google.pdf>
]

[Introduction copied from the White paper]

"The Federated Learning of Cohorts (FLoC) API is a privacy preserving
mechanism proposed within the Chrome Privacy Sandbox, for enabling interest
based advertising. The API is based on the notion of cohorts - groups of
users with similar interests. In this paper we evaluate different methods
for generating cohorts, showing clear trade-offs between privacy and
utility. Using proprietary conversion data, we demonstrate that generating
cohorts based on common interests can significantly improve quality over
random user groupings. In fact, we achieve a 350% improvement in recall and
70% improvement in precision at very high anonymity levels compared to
random user grouping. "

We hope to start to dialog with the broader group on how to improve and
evaluate cohort assignment algorithms.

Thanks,

Deepak Ravichandran

Software Engineer

Google Display Ads
Received on Thursday, 22 October 2020 02:09:46 UTC

This archive was generated by hypermail 2.4.0 : Thursday, 24 March 2022 20:32:27 UTC