Learning to Reweight Examples for Robust Deep Learning

Unofficial PyTorch implementation of Learning to Reweight Examples for Robust Deep Learning.

The paper addresses the problem of imbalanced and noisy datasets by learning a good weighting of examples using a small clean and balanced dataset.

Please Let me know if there are any bugs in my code. Thank you! =)

I implemented this on Python 3.6 using PyTorch 0.4.0.

Dataset

I only ran the experiments for the class imbalance problem. Following the paper, I created an imbalanced training dataset using class '4' and '9' of the MNIST dataset, where '9' is the dominant class. (code for creating the dataset is in data_loader.py)

Note that the test set used to measure the performance is balanced.

Some Results

We can see that even at 0.995 proportion of the dominant class in the training data, the model still reaches 90+% accuracy on the balanced test data.

Acknowledgements

Adrien Ecoffet: https://github.com/AdrienLE