https://github.com/udibr/noisy_labels

TRAINING DEEP NEURAL-NETWORKS USING A NOISE ADAPTATION LAYER
https://github.com/udibr/noisy_labels

Last synced: 2 months ago
JSON representation

TRAINING DEEP NEURAL-NETWORKS USING A NOISE ADAPTATION LAYER

• Host: GitHub
• URL: https://github.com/udibr/noisy_labels
• Owner: udibr
• Created: 2016-11-04T15:17:56.000Z (over 7 years ago)
• Default Branch: master
• Last Pushed: 2017-04-25T21:20:11.000Z (about 7 years ago)
• Last Synced: 2024-01-29T09:35:58.961Z (5 months ago)
• Language: Jupyter Notebook
• Size: 877 KB
• Stars: 118
• Watchers: 6
• Forks: 38
• Open Issues: 4
• Metadata Files:
  • Readme: README.md

Lists

• Awesome-Learning-with-Label-Noise - Noisy Labels with Jupyter Notebook
• Awesome-Learning-with-Label-Noise - Noisy Labels with Jupyter Notebook

README

        
# noisy_labels



TRAINING DEEP NEURAL-NETWORKS USING A NOISE ADAPTATION LAYER

[ICLR 2017 conference submission](http://openreview.net/forum?id=H12GRgcxg)

Learning MNIST when almost half the labels are permuted in a fixed way. For example, when the task of labeling is split between two people that don’t agree.

Follow [mnist-simple](./mnist-simple.ipynb) notebook for an example of how to implement the Simple noise adaption layer in the paper with a single customized Keras layer.

Follow [161103-run-plot](./161103-run-plot.ipynb), [161202-run-plot-cifar100](./161202-run-plot-cifar100.ipynb) and [161230-run-plot-cifar100-sparse](./161230-run-plot-cifar100-sparse.ipynb) notebooks for how to reproduce the results of the paper.