https://github.com/fredhohman/summit-notebooks
Notebooks for Scaling Deep Learning Interpretability by Visualizing Activation and Attribution Summarizations
https://github.com/fredhohman/summit-notebooks
deep-learning deep-learning-visualization interactive-interface interactive-visualization interpretability
Last synced: 12 months ago
JSON representation
Notebooks for Scaling Deep Learning Interpretability by Visualizing Activation and Attribution Summarizations
- Host: GitHub
- URL: https://github.com/fredhohman/summit-notebooks
- Owner: fredhohman
- License: mit
- Created: 2019-03-05T04:53:53.000Z (about 7 years ago)
- Default Branch: master
- Last Pushed: 2019-10-03T01:59:10.000Z (over 6 years ago)
- Last Synced: 2025-03-25T20:06:40.538Z (12 months ago)
- Topics: deep-learning, deep-learning-visualization, interactive-interface, interactive-visualization, interpretability
- Language: Jupyter Notebook
- Homepage:
- Size: 1.4 MB
- Stars: 15
- Watchers: 4
- Forks: 4
- Open Issues: 5
-
Metadata Files:
- Readme: README.md
- License: LICENSE.md
Awesome Lists containing this project
README
# Summit Notebooks
Summit is an interactive system that scalably and systematically summarizes and visualizes what features a deep learning model has learned and how those features interact to make predictions.
This repository contains the python notebooks used to generate the data used in the [Summit visualization][summit].
For the main Summit repo, go to [https://github.com/fredhohman/summit][summit].
### Main notebooks:
* [`activation-matrices.ipynb`](activation-matrices.ipynb): generate aggregated activation matrices
* [`influence.py`](activation-matrices.ipynb): generate aggregated influence matrices
* [`activation-matrices-to-json.ipynb`](activation-matrices-to-json.ipynb): combine activation matrices per class into json format
* [`attribution-graph.ipynb`](dag.ipynb): generating class attribution graphs
* [`feature-vis-sprite-to-images.ipynb`](feature-vis-sprite-to-images.ipynb): split feature visualization sprites to single images
### Experimental notebooks:
* [`top-channels-used-per-layer.ipynb`](top-channels-used-per-layer.ipynb): analysis for determining which channels were used the most by all classes for all layers
## Live Demo
For a live demo, visit: [fredhohman.com/summit][demo]
## Resources
We used the following ImageNet metadata:
* [https://github.com/google/inception/blob/master/synsets.txt](https://github.com/google/inception/blob/master/synsets.txt)
* [https://gist.github.com/aaronpolhamus/964a4411c0906315deb9f4a3723aac57](https://gist.github.com/aaronpolhamus/964a4411c0906315deb9f4a3723aac57)
* [https://gist.github.com/yrevar/942d3a0ac09ec9e5eb3a](https://gist.github.com/yrevar/942d3a0ac09ec9e5eb3a)
## License
MIT License. See [`LICENSE.md`](LICENSE.md).
## Citation
```
@article{hohman2020summit,
title={Summit: Scaling Deep Learning Interpretability by Visualizing Activation and Attribution Summarizations},
author={Hohman, Fred and Park, Haekyu and Robinson, Caleb and Chau, Duen Horng},
journal={IEEE Transactions on Visualization and Computer Graphics (TVCG)},
year={2020},
publisher={IEEE},
url={https://fredhohman.com/summit/}
}
```
## Contact
For questions or support [open an issue][issues] or contact [Fred Hohman][fred].
[summit]: https://github.com/fredhohman/summit
[fred]: https://fredhohman.com
[demo]: https://fredhohman.com/summit/
[issues]: https://github.com/fredhohman/summit-notebooks/issues