Deep Learning on a Healthy Data Diet: Finding Important Examples for Fairness (AAAI 2023)

Mitigating gender bias in text classification models by removing the stereotypical examples (paper). We use the classifiers from 🤗 Hugging Face (link).

How it works

Healthy data diet can be summarized as follows:

1. Generating the counterfactual examples
2. Finding the important examples for fairness using the GE score
3. Adding a pruned version of the original dataset to the important counterfactual examples

Running the experiments

Follow the instructions in the healthy_data_diet_AAAI23.ipynb file to run the experiments in the paper.

Citation

@article{zayed2022deep,
  title={Deep Learning on a Healthy Data Diet: Finding Important Examples for Fairness},
  author={Zayed, Abdelrahman and Parthasarathi, Prasanna and Mordido, Goncalo and Palangi, Hamid and Shabanian, Samira and Chandar, Sarath},
  journal={arXiv preprint arXiv:2211.11109},
  year={2022}
}