This is a Korean review of "Dataset condensation with gradient matching" presented at ICLR 2021. TL;DRDataset Distillation을, 전체 학습 데이터와 소수의 합성 데이터에서 학습된 신경망 가중치의 gradient 간의 일치 문제(gradient matching problem)로 정식화함. Introduction대규모 데이터를 효과적으로 처리하는 전통적인 방법은 coreset construction이며, 이는 *클러스터링 기반의 접근법을 사용함. 또한, continual learning이나 active learning을 통해 대규모 데이터를 효율적으로 다루려는 연구도 활발히 진행되고 있음.이러한 방법들은 일반적으..