Home

Awesome

Noisy_Dropout_regularization

Learning Deep Networks from Noisy Labels with Dropout Regularization

Requirements:

Follow the steps in this order.. Before excuting any of the following, make sure that the modified training set is at place in propoer directory.

Implementation steps:

For all the models:

At test time, remove the noise model and use base model prediction as final prediction.

In case of MNIST dataset on CNN follow the same proceduee as stated above. And for Deep neural network use the given file dnnmnistinit.m and place it into exapmple\mnist folder and perform same set of experiments as for the previous case.

One can find the implementation of randfixedsum.m at this link

Citing:

If you use this work in your research, please cite the following paper:


@inproceedings{jindal2016learning,
  title={Learning deep networks from noisy labels with dropout regularization},
  author={Jindal, Ishan and Nokleby, Matthew and Chen, Xuewen},
  booktitle={Data Mining (ICDM), 2016 IEEE 16th International Conference on},
  pages={967--972},
  year={2016},
  organization={IEEE}
}