A Privacy-Preserving Framework for Collecting Demographic Information
Afra J. Mashhadi · 2020
Currently one of the biggest challenges regarding demographic detection in images and social media is the lack of labelled demographic data. A big part of the challenge is that no suitable mechanism exists to replace traditional intercept surveys in a way that ensures fairness and inclusion. The lack of labelled data has also impacted the training of AI algorithms. That is the lack of labels relevant to the target domains has made it hard to estimate the accuracy of the AI algorithms when they are applied to real world situations. In this paper, we propose a framework for collecting in-the-wild images and demographic labels from ordinary people (e.g., park visitors) that ensures that privacy is integrated at every stage of the data collection process from storage to processing and sharing.