FAIR Data

FAIR (Findable, Accessible, Interoperable, Reusable) data principles are being extended to encompass AI models and datasets, aiming to improve the trustworthiness and ethical implications of machine learning. Current research focuses on mitigating bias in datasets through techniques like data removal, fair sampling in generative models (e.g., diffusion models), and developing fair mapping methods to transform data distributions while preserving utility. This work is crucial for ensuring the reliability and fairness of AI systems across various scientific domains and practical applications, promoting responsible data usage and preventing discrimination.