single source · data curation
Random selection among language annotations during DROID training
Randomly selecting among the natural-language annotations supplied for each training episode.
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 we randomize over them during training, randomize over them during training.
Filed under data curation › data mixture & curriculum