D3M: Improving Group Robustness via Dataset Selection
Paper Code Machine learning models are increasingly making decisions in high-stakes scenarios, from healthcare to finance to criminal justice. These models are trained on large-scale datasets that often contain biased data. As a result, these models often exhibit disparate performance across different subgroups of the data. For instance, facial recognition systems have been shown to perform poorly on images of Black women, while medical imaging models struggle with X-rays of patients without chest drains. Such […]
Like
0
Liked
Liked