Women Also Snow Board

Image classifiers were shown images of people snow boarding, they were much more likely to label the person as il. When you trained models on thi data, it magnified an id this bias. And so whereas at train time, maybe you had a skew of 70, 30, men, women snow boarding, now at test time you have 90, ten. Once they recognized the image classifi as were biased, lisa anne and her colleagues shifted the garithm to focus on the subject of the photograph rather than their surroundings.

Transcript

Play full episode

The AI-powered Podcast Player

Save insights by tapping your headphones, chat with episodes, discover the best highlights - and more!

Get the app