mixup: Beyond Empirical Risk Minimization

ICLR 2018 Hongyi ZhangMoustapha CisseYann N. DauphinDavid Lopez-Paz

Large deep neural networks are powerful, but exhibit undesirable behaviors such as memorization and sensitivity to adversarial examples. In this work, we propose mixup, a simple learning principle to alleviate these issues... (read more)

PDF Abstract
TASK DATASET MODEL METRIC NAME METRIC VALUE GLOBAL RANK RESULT BENCHMARK
Image Classification CIFAR-10 DenseNet-BC-190 + Mixup Percentage correct 97.3 # 22
Image Classification CIFAR-100 DenseNet-BC-190 + Mixup Percentage correct 83.20 # 17
Semi-Supervised Image Classification CIFAR-10, 250 Labels MixUp Accuracy 52.57 # 8
Domain Generalization ImageNet-A Mixup (ResNet-50) Top-1 accuracy % 6.6 # 3
Image Classification Kuzushiji-MNIST PreActResNet-18 + Input Mixup Accuracy 98.41 # 6
Semi-Supervised Image Classification SVHN, 250 Labels MixUp Accuracy 60.03 # 9

Methods used in the Paper