Our toolkit
The MAI-BIAS toolkit is a low-code fairness assessment and bias mitigation app. It contains 40+ modules from our partners and third parties, accessible through a standardized interface. It adjusts to your setting with out-of-the-box support for many data types, models, and fairness concerns.
How to ...
Everyone can explore results, including policymakers and non-experts. Examples:
Toolkit gallery
Click on any screenshot to enlarge it.
Want to add your own modules or report an issue? Visit the GitHub repositories of the local runner or server runner.
Reuse our software
Apply bias detection and mitigation code in your projects independently. Software listed here also supports the MAI-BIAS toolkit.
FairBenchAny dataLibrary
A comprehensive AI fairness exploration library with >300 standardized measures. It can parse any data modality, including generated text.
MMM-fairAny dataLibrary
A library that supports high-stakes AI decision-making under competing fairness and accuracy demands.
FairBranchAny dataAlgorithm
Implementation of FairBranch pipelines for vision and tabular modalities.
FB-tinyTabular dataTerminal
Terminal tool for bias audit. Vs one FairBench or AIF360 run:
x30 faster
x17 less energy
x50 less memory
VB-MitigatorVisionAlgorithm
Implementation and evaluation of existing and new visual bias mitigation methods.
SDFDVisionDataset
A synthetic face image dataset that captures broad facial diversity (demographics, biometrics, non-permanent traits).
HyperFairNetworksLibrary
A Python library for generating, evaluating, and improving rankings under fairness constraints.
