IDnow is an expert in remote identity verification and commercializes the product IDCheck.io to several companies to help them with their KYC (Know Your Customer) process. IDnow is also involved in the development of a digital identity at the national level with its participation in France Connect, and at the European level with the project SOTERIA, aiming to develop a personal data management platform coupled with a strong and secure digital identity. These digital identity solutions are based on the product IDCheck for the identity verification part.
One fundamental task for identity verification is the face comparison between the person on the ID document and the person behind the screen through a submitted self-portrait. IDnow has developed its own face recognition algorithm, specifically adapted for identity verification. Despite its global accuracy in this specialized context, it has been observed that the obtained results are biased for some specific cases such as bad quality mobile captures or dark-skinned faces captures. Many false rejections were due to missing faces detection. An initial step to reduce the rejections was to train an in-house face detector for this aim. Unfortunately, the observed behavior of the face verification for the newly detected faces was biased. The goal of this position is to perform research work on bias detection and mitigation approaches to improve the hidden biases of the system allowing fair access to the service.
The candidate will conduct his research activity in collaboration with international academic specialists in the field (CERTH, UniBw, EXUS) and will have the opportunity to publish his research work in international conferences. The work is funded by the European project MAMMOth whose aim is to promote diversity and inclusion in the design, development and deployment of Artificial Intelligence systems, by both building the capacity of relevant stakeholders and by providing bias-preventing AI solutions into an open-source suite.
Tasks
Contribution in the construction of a high-quality dataset for Deep learning tasks
Collaboration with the project partners on the research related to bias mitigation
Evaluation and integration of the MAMMOth toolbox for bias evaluation and mitigation
Project task management and interaction with the project partners
Regular presentation of the work advance with the project consortium
Participation in the writing of scientific papers and research reports
Bilateral collaboration with the other research teams on related research topics
Keywords : Computer Vision, deep Learning, Python, Biometrics, bias, face verification
A doctoral degree (PhD) in Artificial Intelligence, Computer Vision or related fields.
Verified experience in AI and/or image processing (in the academia or industry)
Ability to write robust code in Python
ML and data processing libraries (Pytorch, pandas, ML Flow, jupyter, … )
Good problem-solving skills
Excellent English skills for communication and writing scientific articles
Duration:
1 year and a half