2024
journal
Bt-GAN: Generating Fair Synthetic Healthdata via Bias-transforming Generative Adversarial Networks
Resmi Ramachandranpillai, Md Fahim Sikder, David Bergström, Fredrik Heintz
Journal of Artificial Intelligence Research
Theme · bias
Bias, Representation & Intersectionality
Synthetic data inherits — and amplifies — the biases of the data and models that produce it. We study representation gaps and intersectional effects in generated datasets.
Projects
active ontology · bias · transparency
Synthetic Data — Facts, Representations, and Transparency
A five-year WASP-HS research environment examining how AI-generated synthetic data is produced, deployed, and contested — and what it means for the people and institutions that come to rely on it.
completed bias · hallucinations
Social Complexity and Fairness in Synthetic Medical Data
A WASP-HS NetX project examining how machine learning-generated synthetic medical datasets distort representation. Using intersectional analysis, the project developed fairness metrics for synthetic medical data, exploring how intersectionality can contribute to better, more equitable data.