Chameleon

Chameleon is a fairness-aware data augmentation method that leverages foundation models to enhance coverage of minorities in multimodal datasets. This project addresses critical challenges in algorithmic fairness and data representation.

Key Achievements

• 22% improvement in model accuracy on under-represented groups on the FERETDB benchmark
• Novel approach to fairness-aware data augmentation using foundation models
• Published in Proceedings of the VLDB Endowment (2024)

Technical Innovation

Chameleon employs advanced foundation models to generate synthetic data that specifically targets underrepresented groups, ensuring more balanced and fair training datasets for machine learning models.

Research Impact

This work contributes to the growing field of responsible AI by providing practical solutions for addressing bias in training data, particularly in computer vision applications where demographic representation is crucial.

Publication Details

This research was published in the Proceedings of the VLDB Endowment and represents a significant contribution to the field of responsible data management and algorithmic fairness.