Open-source library underpinning fairness assessment in Azure Machine Learning, offering disparity metrics plus reduction and post-processing mitigation algorithms.
Read more: Microsoft Learn
In the Ultra Transcenders books
Each book explains Fairlearn in context, with comparison tables and the common traps.
Terms in this definition
- Fairness assessment
Responsible AI dashboard tool, based on Fairlearn, that measures gaps in model performance and selection rate between groups formed from sensitive features like age, ethnicity or gender.
- Azure Machine Learning
Azure platform where you train, deploy, monitor and retrain models of your own, practising MLOps through tools such as pipelines, online endpoints, model monitoring and prompt flow.
Related terms
- CorrelationRemover
Applied before training, this Fairlearn transformer cancels out the correlation that other features share with sensitive ones; you then need to train the model again.
- ExponentiatedGradient
A Fairlearn reduction method: it wraps a binary classification estimator and repeatedly retrains it on reweighted data so the result meets equalized odds or demographic parity.
- GridSearch (Fairlearn)
A reduction algorithm in Fairlearn that retrains an estimator across a grid of different reweightings. It returns multiple models so you can pick the balance between accuracy and disparity.
- ThresholdOptimizer
A post-processing algorithm in Fairlearn that sets separate thresholds per group on an already-trained binary classifier, targeting equalized odds or demographic parity with no retraining.