Appendix

From the Appendix A.1, we obtain the gradient of the sample-wise Source code for the experiments is available in the zip file. All test accuracy are recorded from the last epoch of training. For Clothing1M, it provides 50k, 14k, 10k refined clean data for training, validation and testing respectively. Note that we do not use the 50k clean data for fair comparison with existing methods. The information of datasets are described in Table 1.