Mitigating Bias in Dataset Distillation