Captum

Captum is a model interpretability and understanding library for PyTorch. Captum means comprehension in Latin and contains general purpose implementations of integrated gradients, saliency maps, smoothgrad, vargrad and others for PyTorch models. It has quick integration for models built with domain-specific libraries such as torchvision, torchtext, and others.

Captum is integrated in the Kaa Explainability Platform

Documentation

Methodological Guidelines

• Methodological Guideline for Explainability

Scientific Contribution

• Comments on ISO 6254 (Explainability)

Benchmark

• Explicability benchmark

Application over use-case

• Unsupervised Anomaly Detection and Explainability tools for Time Series

State of the Art

• Characterization of the notion of trust

Support

Support for Captum must be obtained directly from the component owners, you can contact them directly here.