Co-teaching

by bhanML

bhanML / Co-teaching

NeurIPS'18: Co-teaching: Robust Training of Deep Neural Networks with Extremely Noisy Labels

241 Stars 63 Forks Last release: Not found 35 Commits 0 Releases

Available items

No Items, yet!

The developer of this repository has not created any items for sale yet. Need a bug fixed? Help with integration? A different license? Create a request here:

Co-teaching

NeurIPS'18: Co-teaching: Robust Training of Deep Neural Networks with Extremely Noisy Labels (Pytorch implementation).

Another related work in NeurIPS'18:

Masking: A New Perspective of Noisy Supervision

Code available: https://github.com/bhanML/Masking

========

This is the code for the paper: Co-teaching: Robust Training of Deep Neural Networks with Extremely Noisy Labels
Bo Han, Quanming Yao, Xingrui Yu, Gang Niu, Miao Xu, Weihua Hu, Ivor Tsang, Masashi Sugiyama
To be presented at NeurIPS 2018.

If you find this code useful in your research then please cite

bash
@inproceedings{han2018coteaching,
  title={Co-teaching: Robust training of deep neural networks with extremely noisy labels},
  author={Han, Bo and Yao, Quanming and Yu, Xingrui and Niu, Gang and Xu, Miao and Hu, Weihua and Tsang, Ivor and Sugiyama, Masashi},
  booktitle={NeurIPS},
  pages={8535--8545},
  year={2018}
}

Setups

All code was developed and tested on a single machine equiped with a NVIDIA K80 GPU. The environment is as bellow:

  • CentOS 7.2
  • CUDA 8.0
  • Python 2.7.12 (Anaconda 4.1.1 64 bit)
  • PyTorch 0.3.0.post4
  • numpy 1.14.2

Install PyTorch via:

bash
pip install http://download.pytorch.org/whl/cu80/torch-0.3.0.post4-cp27-cp27mu-linux_x86_64.whl

Running Co-teaching on benchmark datasets (MNIST, CIFAR-10 and CIFAR-100)

Here is an example:

python main.py --dataset cifar10 --noise_type symmetric --noise_rate 0.5 

Performance

| (Flipping, Rate) | MNIST | CIFAR-10 | CIFAR-100 | | ---------------: | -----: | -------: | --------: | | (Pair, 45%) | 87.58% | 72.85% | 34.40% | | (Symmetry, 50%) | 91.68% | 74.49% | 41.23% | | (Symmetry, 20%) | 97.71% | 82.18% | 54.36% |

Contact: Xingrui Yu ([email protected]); Bo Han ([email protected]).

AutoML

Please check the automated machine learning (AutoML) version of Co-teaching in - Searching to Exploit Memorization Effect in Learning from Corrupted Labels. ICML-2020 paper code

We use cookies. If you continue to browse the site, you agree to the use of cookies. For more information on our use of cookies please see our Privacy Policy.