Neily Sanon – L’identification et la mesure des biais dans les systèmes de reconnaissance facial

Facial recognition systems exhibit varying levels of performance depending on demographic characteristics such as gender, age, or ethnic origin. These biases, extensively documented in empirical studies, raise critical concerns regarding reliability, fairness, and ethical compliance.

This presentation will examine the underlying sources of these biases, whether they stem from dataset composition, algorithmic choices, or intersectional factors. It will also explore the different evaluation methods used to quantify performance disparities across groups, along with mitigation strategies that can be applied at various stages of the processing pipeline Particular attention will be given to emerging certification mechanisms, as regulatory authorities begin to incorporate bias evaluation into their assessment and approval processes.