Researchers show that computer vision algorithms pretrained on ImageNet exhibit multiple, distressing biases

Researchers say they've found evidence of pervasive bias in computer vision algorithms trained on ImageNet, a popular photo dataset.