vision-and-language models

This article investigates the unwanted bias in pretrained vision-and-language (V&L) models, with a focus on gender bias. The authors quantified bias amplification in pretraining and after fine-tuning on three families…