IEEE Transactions on Radiation and Plasma Medical Sciences 2021

Interpretability Survey

On Interpretability of Artificial Neural Networks: A Survey

Fenglei Fan, Jinjun Xiong, Mengzhou Li, Ge Wang

Current or former lab member. Select a name to see all of their projects.

Method overview of Interpretability Survey
Taxonomy used for the interpretability review (Fig. 2 of the paper).

Abstract

Deep neural networks succeed with text, images, video and graphs, but their black-box nature holds back their use in high-stakes settings such as diagnosis and therapy. This survey proposes a taxonomy of interpretability, reviews recent work on improving it, describes applications in medicine, and discusses future directions, including links to fuzzy logic and brain science.

The repository holds the authors’ independent evaluation of several existing interpretability methods, which accompanies the survey.

What’s in the repository

  • Code for the evaluation figures in the survey, one folder per figure
  • A link to the complete files, including a saved model that is too large for the repository

Citation

@article{fan2021interpretability,
  title   = {On Interpretability of Artificial Neural Networks: A Survey},
  author  = {Fan, Fenglei and Xiong, Jinjun and Li, Mengzhou and Wang, Ge},
  journal = {IEEE Transactions on Radiation and Plasma Medical Sciences},
  volume  = {5},
  number  = {6},
  pages   = {741--760},
  year    = {2021},
  doi     = {10.1109/TRPMS.2021.3066428}
}
Back to projects