Differential Privacy is a state-of-the-art Privacy Enhancing Technology which allows one to gather aggregated information without risking individual's privacy, however it comes at the cost of a privacy accuracy trade-off. Recently there have been some publications which expose a disparate accuracy loss over different subgroups of the training dataset when implementing a private version of Stochastic Gradient Descent.
This showcases how focusing and optimizing specific factors (Privacy), can have negative impacts on others ("Fairness"). A system's thinking approach allows one to deal with such complexities as well as the social and contextual factors of a system.
In my work I aim to not just have a better understanding of the causes of disparate accuracy drop when implementing Differential Privacy but also have a better understanding of the workings of the Consumer Credit Industry and their tech implementation process. The combination of the more traditional technical study on Differential Privacy as well as the analysis of the Consumer Credit Industry from a systems perspective will allow one to study the long and short-term impact of Differential Privacy within this context, from the perspective of the several stakeholders.
I am currently recruiting for my next study: UK Consumer Credit Stakeholder Consultation
The study is made up of a 60~90 minute interview (held online).
It aims to understand the workings of the Industry and Interactions between staleholders as well as expert's attitudes towards Differential Privacy.
Participants will not have to disclose specific commercial information but are require line manager approval.
I am looking to recruit people who are, over the age of 18 who work or have worked in the Consumer Credit Industry, correspondent regulatory bodies (e.g. FCA, ICO) or work for credit/debt charities ans consumer advocate groups.
Participation in the study will be compensated with a £15 voucher.
If you are interested in the study don't hesitate in contacting me through email.
This author is supported by the Horizon Centre for Doctoral Training at the University of Nottingham (UKRI Grant No. EP/S023305/1).