Dataset Distillation by Influence Matching
This paper proposes a new method for dataset distillation that focuses on aligning the final outcome of training, rather than imitating the training process. Practitioners might care about this method because it can improve the accuracy of model training on limited data.