1 paper · 1 filter
Xuhui Li, Zhengquan Luo, Xiwei Liu +2
Dataset distillation compresses large datasets into compact synthetic sets with comparable performance in training models. Despite recent progress on diffusion-based distillation,…