training_with_random_labels

This notebook replicates the results from "Understanding deep learning requires rethinking generalization".

@article{zhang2016understanding,
  title={Understanding deep learning requires rethinking generalization},
  author={Zhang, Chiyuan and Bengio, Samy and Hardt, Moritz and Recht, Benjamin and Vinyals, Oriol},
  journal={arXiv preprint arXiv:1611.03530},
  year={2016}
}

In this notebook a Resnet model is trained on the CIFAR10 dataset. The tranining is done on cifar10 dataset with option for random labels.

The neural network model was taken from https://github.com/xternalz/WideResNet-pytorch.

The noetbook is inspired by https://github.com/pluskid/fitting-random-labels.

The notebook was developed and tested on https://colab.research.google.com/ with GPU backend.