Keywords:	face recognition, machine learning, fairness, open set learning, novelty detection, out-of-distribution data
Prerequisites:	Uncertainty Quantification in Machine Learning, Python, basic statistics
Difficulty:	Hard (B.Sc.), Medium (M.Sc.—see notes below)

Abstract

Face recognition (FR) is a technology which has been around since the early days of computer vision and comprises several Computer Vision tasks related to identifying, localizing, verifying the identity, etc. of faces in images. Due to dealing with strictly personal data, and often also being utilized in the context of law enforcement, FR has been the subject of several ethical and legal debates, and has also been dedicated a full study on the recently approved AI Act by the European Parliament. FR has, moreover, been the subject of several studies on the topic of fairness, like the notable Gender Shades project, which have identified traces of racial and age-related biases in some commercially-available FR models, likely due to the lack of diversity in the training data. There are several works coupling FR with the concept of Open-Set Recognition (OSR), whereas the models trained can adapt to the presence of unknown identities in the test set. However, often these works do not consider possible bias in their implementations, thus possibly causing the deployment of a potentially discriminatory model.

Required work

• Literature review on the concept of fairness in machine learning, with a focus on FR
• Literature review on the concept of OSR in FR
• Pick one or two models with open implementations and evaluate their fairness on one or more benchmark datasets
• (Optional for B.Sc. students) Use methods for bias correction to improve the fairness of the models
• (Possible extensions) Uncertainty quantification analysis on models, e.g., calibration, reliability diagrams, etc.

Relevant literature

• Survey on fairness: Mehrabi et al. A Survey on Bias and Fairness in Machine Learning. ACM Computing Surveys. 2021.
• Ways to measure fairness in FR: Howard et al. Evaluating Proposed Fairness Models for Face Recognition Algorithms. 2022.
• Recent work on OSR + FR: Vareto et al. Open-set Face Recognition using Ensembles trained on Clustered Data. 2023.
• Recently-published dataset for fairness assessment: Gustafson et al. FACET: Benchmarking fairness of vision models. 2023.