Sequential Subset Matching for Dataset Distillation

The synthetic datasets are expected to capture the essence of the knowledge contained in real-world datasets such that the former yields a similar performance as the latter.