We propose FairImagen that debiases text-to-image models by post-processing prompt embeddings, improving fairness across gender and race without retraining the model.

We developed a functional autoencoder framework for cluster analysis of multi-dimensional functional data, data that are random realizations of vector-valued random functions.