Idanwo ti o sopọ mọ itọsọna · Alabọde Ipele

Label Bias and Annotator Disagreement Quiz

Examine subjective labels, annotator identity and instructions, proxy targets, disagreement, and safer ways to preserve uncertainty in training data.

Jẹmọ itọsọna awọn ọnaLabel Bias And Annotator Disagreement
Ibeere 1 ti 8

A model learns “toxic” labels that reflect annotators’ assumptions about dialect. What risk is most direct?