As the state-of-the-art machine learning methods in many fields rely on
larger datasets, storing datasets and training models on them become
significantly more expensive. This paper proposes a training set synthesis
technique for data-efficient learning, called dataset condensation, th