Code for paper Learning to Reweight Examples for Robust Deep Learning. [arxiv]

We tested the code on

• tensorflow 1.10
• python 3

Other dependencies:

• numpy
• tqdm
• six
• protobuf

The following command makes the protobuf configurations.

make

python -m mnist.mnist_train --exp ours

Please see mnist/mnist_train.py for more options.

bash cifar/download_cifar.sh ./data

Config files are located in cifar/configs. For ResNet-32, use cifar/configs/cifar-resnet-32.prototxt. For Wide ResNet-28, use cifar/configs/cifar-wide-resnet-28-10.prototxt.

python -m cifar.cifar_train --config [CONFIG]

Please see cifar/cifar_train.py for more options.

python -m cifar.cifar_train_background --config [CONFIG]

Please see cifar/cifar_train_background.py for more options.

If you use our code, please consider cite the following: Mengye Ren, Wenyuan Zeng, Bin Yang, Raquel Urtasun. Learning to Reweight Examples for Robust Deep Learning. ICML 2018.

@inproceedings{ren18l2rw,
  author    = {Mengye Ren and Wenyuan Zeng and Bin Yang and Raquel Urtasun},
  title     = {Learning to Reweight Examples for Robust Deep Learning},
  booktitle = {ICML},
  year      = {2018},
}