AI Explainability 360


This extensible open source toolkit can help you comprehend how machine learning models predict labels by various means throughout the AI application lifecycle. We invite you to use it and improve it.

 

Not sure what to do first? Start here!

Read More

Learn more about explainability concepts, terminology, and tools before you begin.


*

Try a Web Demo

Step through the process of explaining models to consumers with different personas in an interactive web demo that shows a sample of capabilities available in this toolkit.


*

Watch Videos

Watch videos to learn more about AI Explainability 360 toolkit.


*

Read a Paper

Read a paper describing how we designed AI Explainability 360 toolkit.


*

Use Tutorials

Step through a set of in-depth examples that introduce developers to code that explains data and models in different industry and application domains.


*

Ask a Question

Join our AI Explainability 360 Slack Channel to ask questions, make comments, and tell stories about how you use the toolkit.


*

View Notebooks

Open a directory of Jupyter notebooks in GitHub that provide working examples of explainability in sample datasets. Then share your own notebooks!


*

Contribute

You can add new algorithms and metrics in GitHub. Share Jupyter notebooks showcasing how you have enabled explanations in your machine learning application.


*

Learn how to put this toolkit to work for your application or industry problem. Try these tutorials.

Credit Approval

See how to explain credit approval models using the FICO Explainable Machine Learning Challenge dataset.


*

Medical Expenditure

See how to create interpretable machine learning models in a care management scenario using Medical Expenditure Panel Survey data.


*

Dermoscopy

See how to explain dermoscopic image datasets used to train machine learning models that help physicians diagnose skin diseases.


*

Health and Nutrition Survey

See how to quickly understand the National Health and Nutrition Examination Survey datasets to hasten research in epidemiology and health policy.


*

Proactive Retention

See how to explain predictions of a model that recommends employees for retention actions from a synthesized human resources dataset.


*

These are eight state-of-the-art explainability algorithms that can add transparency throughout AI systems. Add more!

Boolean Decision Rules via Column Generation (Light Edition)

Directly learn accurate and interpretable ‘or’-of-‘and’ logical classification rules.


*

Generalized Linear Rule Models

Directly learn accurate and interpretable weighted combinations of ‘and’ rules for classification or regression.


*

ProfWeight

Improve the accuracy of a directly interpretable model such as a decision tree using the confidence profile of a neural network.


*

Teaching AI to Explain its Decisions

Predict both labels and explanations with a model whose training set contains features, labels, and explanations.


*

Contrastive Explanations Method

Generate justifications for neural network classifications by highlighting minimally sufficient features, and minimally and critically absent features.


*

Contrastive Explanations Method with Monotonic Attribute Functions

Contrastive explanations for colored images or images with rich structure.


*

Disentangled Inferred Prior VAE

Learn disentangled representations for interpreting unlabeled data.


*

ProtoDash

Select prototypical examples from a dataset.


*

Although it is ultimately the consumer who determines the quality of an explanation, the research community has proposed quantitative metrics as proxies for explainability.

Faithfulness

Correlation between the feature importance assigned by the interpretability algorithm and the effect of features on model accuracy.


*

Monotonicity

Test whether model accuracy increases as features are added in order of their importance.


*

About this site

AI Explainability 360 was created by IBM Research and donated by IBM to the Linux Foundation AI & Data.

Additional research sites that advance other aspects of Trusted AI include:

AI Fairness 360
AI Privacy 360
Adversarial Robustness 360
Uncertainty Quantification 360
AI FactSheets 360