Refreshing Less, Selecting Better: Reusing Stale Gradient Features for Efficient Influence-Based Data Selection
Cached Diverse Influence Selection (CDIS), which recomputes gradient features for the top-ranked fraction of candidates under the stale scores, fits an affine calibration on the recomputed examples, and selects the final subset under source and length quotas.
Jian-Chang Su, Yi-Fan Zhang, Wei Zhang
· 0 citations