hendrycks / error-detection

A Baseline for Detecting Misclassified and Out-of-Distribution Examples in Neural Networks

A Baseline for Detecting Misclassified and Out-of-Distribution Examples in Neural Networks

Implementations of the experiments found in A Baseline for Detecting Misclassified and Out-of-Distribution Examples in Neural Networks by Dan Hendrycks and Kevin Gimpel. https://arxiv.org/abs/1610.02136

Most results are in Jupyter notebooks since several data sets used have licensing restrictions (e.g., TIMIT, WSJ PTB, etc.).

Citation

@article{hendrycks17baseline,
  author    = {Dan Hendrycks and Kevin Gimpel},
  title     = {A Baseline for Detecting Misclassified and Out-of-Distribution Examples in Neural Networks},
  journal = {Proceedings of International Conference on Learning Representations},
  year = {2017},
}

Follow-up Project

This more recent repository has PyTorch code for a general anomaly detection method which builds on this baseline.