Skip to content

Resources we like for Explainable Artificial Intelligence: sites, papers, implementations, and more!

Notifications You must be signed in to change notification settings

SFI-Visual-Intelligence/awesome-xai

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

7 Commits
 
 

Repository files navigation

awesome-xai

Resources we like for Explainable Artificial Intelligence: sites, papers, implementations, and more!

This page is set up as part of the Interpretable Learning Spring 2022 curriculum arranged at UiT The Arctic University, but it is open to everyone who wants to contribute - send a pull request if you feel something is missing!

Primer on explainable artificial intelligence

Briefly, explainable artificial intelligence (XAI) methods try to create machine learning models capable of explaining how they arrive at predictions. We can use these methods to verify that the models are using features relevant to prediction, to identify areas where the models need improvement, and to extend our own understanding of the problems the models solve.

We separate between

  • Intrinsic and post-hoc explanations: intrinsically interpretable models rely on the model design being inherently understandable, while post-hoc methods try to infer conclusions about the model after training
  • Model-specific and model-agnostic explanations: model-specific methods rely on certain properties of the model design to provide explanations, while model-agnostic methods can be adapted for all models
  • Local and global explanations: local explanations attempt to explain specific predictions for a limited set of data points, while global explanations show patterns and rules applied for every input point

For a longer introduction, Nirmal Sobha Kartha's article for The Gradient, Explain Yourself - A Primer on ML Interpretability & Explainability, is highly recommended.

Libraries and repositories

  • pair-code/what-if-tool
    • Tool for exploring black-box classification/regression models - partial dependence plots and
  • EthicalML/xai
    • Python library with utilities for showing per-group statistical metrics, plotting feature importance, upsampling/downsampling to balance a dataset against specific attributes
  • Quantus
    • Toolkit for quantitative evaluation of explanation methods

Interpretable models

Post-hoc explanations

Books

  • Interpretable Machine Learning (2022) by Christoph Molnar
    • Available freely online, available for purchase in PDF/ebook format on Leanpub and in print on lulu.com
  • Interpretable AI by Ajay Thampi
    • Part of Manning's MEAP series, publishes in May 2022

Courses / MOOCs

Articles

More recent articles

Poly-cam high resolution class activation map for convolutional neural networks (2022) by Englebert. Combine the activations of different layers, upsampling the last layers, to produce high resolution calss activation maps.

About

Resources we like for Explainable Artificial Intelligence: sites, papers, implementations, and more!

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published