Analysing the Noise Model Error for Realistic Noisy Label Data

Additional material for the publication

Hedderich, Zhu and Klakow:

Analysing the Noise Model Error for Realistic Noisy Label Data

AAAI 2021

https://ojs.aaai.org/index.php/AAAI/article/view/16938

Structure

This additional material is split into three parts:

• NoisyNER: You can find our newly proposed dataset for evaluating noisy-label settings in this separate repository https://github.com/uds-lsv/NoisyNER
• Noise Estimation Experiments: The code for the experiments comparing the theoretical, expected noise model error to the empirical measurements can be found in the subdirectory exp_noise_model_error.
• Base Model Performance: The code for the experiments showing the relationship between noise estimation and base model performance can be found in the subdirectory exp_base_model_performance.

Please refer to the README files in each directory for additional information on installation, reproduction, etc.

Contact & Citations

For more details, please refer to our publication https://arxiv.org/abs/2101.09763. If you have any questions or if you run into any issues, feel free to contact us.

When you work with this dataset, please consider citing us as

@inproceedings{hedderich2021analysing,
  title={Analysing the Noise Model Error for Realistic Noisy Label Data},
  author={Hedderich, Michael A and Zhu, Dawei and Klakow, Dietrich},
  booktitle={Proceedings of the AAAI Conference on Artificial Intelligence},
  volume={35},
  number={9},
  pages={7675--7684},
  year={2021}
}