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

Learn about groundbreaking research in facial recognition bias mitigation through this 41-minute seminar presentation from AutoML Seminars. Explore how neural network architectures themselves contribute to bias in face recognition systems, challenging the conventional belief that biases stem primarily from training data. Discover innovative approaches using neural architecture search and hyperparameter optimization to develop fairer facial recognition models that outperform existing bias mitigation methods. Examine the practical applications and results across major datasets like CelebA and VGGFace2, while understanding how these principles extend to tabular data analysis. Access the complete research paper and implementation code to dive deeper into this novel approach for creating more equitable AI systems.