DNN-DataDependentActivation

This repository consists PyTorch code for deep neural networks with graph interpolating function as output activation function

External dependency: pyflann (https://github.com/primetang/pyflann)

Place the pyflann library in your current directory to replace the pyflann folder

Cifar10-Natural

Code for reproducing results of naturally trained ResNets on the Cifar10

Cifar10-Robust

Code for reproducing results of PGD adversarial training for ResNets on the Cifar10

MNIST-Robust

Code for reproducing results of PGD adversarial training for Small-CNN on the MNIST

If you find this work useful and use it on you own research, please cite our paper

@incollection{NIPS2018_7355,
title = {Deep Neural Nets with Interpolating Function as Output Activation},
author = {Wang, Bao and Luo, Xiyang and Li, Zhen and Zhu, Wei and Shi, Zuoqiang and Osher, Stanley},
booktitle = {Advances in Neural Information Processing Systems 31},
editor = {S. Bengio and H. Wallach and H. Larochelle and K. Grauman and N. Cesa-Bianchi and R. Garnett},
pages = {743--753},
year = {2018},
publisher = {Curran Associates, Inc.},
url = {http://papers.nips.cc/paper/7355-deep-neural-nets-with-interpolating-function-as-output-activation.pdf}
}

And the longer version is available at

@ARTICLE{Wang:2019Interpolation,
       author = {B. Wang and S. Osher},
        title = "{Graph Interpolating Activation Improves Both Natural and Robust Accuracies in Data-Efficient Deep Learning}",
      journal = {arXiv e-prints},
         year = "2019",
        month = "Jul",
          eid = {arXiv:1907.06800},
        pages = {arXiv:1907.06800},
archivePrefix = {arXiv},
       eprint = {},
 primaryClass = {stat.ML}
}

Dependence

PyTorch 0.4.1