Skip to content

Captum

  • Publisher: MetaAI
  • Status: active
  • Version: 0.8
  • Release Date: 2019-10-DD
  • Date Added: 2026-01-13
  • Source URL: Link to official repo/page

Summary

Captum is a Pytorch Library that can identify the specific features and attributes that are affecting the accuracy of the result. Captum has different analysis tools that can be utilized such as GradientShap, DeepLift, IntegratedGradients, LayerConductance, NeuronConductance, and NoiseTunnel. Each of these identifies accuaracy in a unique way and allows the user to test different codes in a variety of ways.

Key Takeaways

  • The Integrated Gradients feature can be used to analyze important words in a text model output.
  • Can pinpoint the accuracy of certian features.
  • Can use regression models on datasets to understand networks with improved accuracy

code/security-tools/captum

Additional Sources

  • Captum Website — has tutorials, documents, and references for all the unique features and API
  • Captum Version Directory — describes and provides dates for all the updates applied to Captum from 2019 to present time.

License

BSD 3-Clause