Dataset Condensation with Differentiable Siamese Augmentation

                          Bo Zhao 1 Hakan Bilen 1

             Abstract                 massive datasets (e.g. (Kuznetsova et al., 2020; Chen et al.,
   In many machine learning problems, large-scale        2020a; Kwiatkowski et al., 2019)). For example, it takes
   datasets have become the de-facto standard to         12.3k TPU days to train EfficientNet-L2 (Xie et al., 2020) on
   train state-of-the-art deep networks at the price  ...