Rhea Sukthanker and Samuel Dooley - Rethinking Bias Mitigation

Title: Rethinking Bias Mitigation: Fairer Architectures Make for Fairer Face-Recognition

Abstract:
Face recognition systems are widely deployed in safety-critical applications, in- cluding law enforcement, yet they exhibit bias across a range of socio-demographic dimensions, such as gender and race. Conventional wisdom dictates that model biases arise from biased training data. As a consequence, previous works on bias mitigation largely focused on pre-processing the training data, adding penalties to prevent bias from effecting the model during training, or post-processing predictions to debias them, yet these approaches have shown limited success on hard problems such as face recognition. In our work, we discover that biases are actually inherent to neural network architectures themselves. Following this reframing, we conduct the first neural architecture search for fairness, jointly with a search for hyperparameters. Our search outputs a suite of models which Pareto-dominate all other high-performance architectures and existing bias mitigation methods in terms of accuracy and fairness, often by large margins, on the two most widely used datasets for face identification, CelebA and VGGFace2. Furthermore, these models generalize to other datasets and sensitive attributes. Finally we show that these ideas of fairness through NAS+HPO carryover to tabular data.

Speaker: Rhea Sukthanker and Samuel Dooley

Paper: https://arxiv.org/abs/2210.09943

Code: https://github.com/dooleys/FR-NAS