1 paper · 1 filter
Ruonan Yu, Songhua Liu, Zigeng Chen +2
Dataset distillation or condensation aims to condense a large-scale training dataset into a much smaller synthetic one such that the training performance of distilled and original…