Open Models Almanac
Methodsdata curationdata mixture & curriculum

single mention · data curation

Training-data bias mitigation through reweighting and augmentation

Mitigating data bias by upsampling selected language data and adding translated or synthetically rephrased examples.

The sources on this page only mention it in passing. The line above is the extraction's one-line gloss. See data mixture & curriculum for the methods filed beside it.

Also called Bias mitigation.

Filed under data curation › data mixture & curriculum