Trusted-AI / AIF360

A comprehensive set of fairness metrics for datasets and machine learning models, explanations for these metrics, and algorithms to mitigate bias in datasets and models.
https://aif360.res.ibm.com/
Apache License 2.0
2.44k stars 835 forks source link

Mitigate bias in SVM classifier using AdversarialDebiasing method #437

Open pradeepdev-1995 opened 1 year ago

pradeepdev-1995 commented 1 year ago

How to mitigate bias in the SVC classifier using AdversarialDebiasing (in-process bias mitigation) method? In the documentation, it seems that it learns its own classifier. So how to load our own specific classifier like SVM to AdversarialDebiasing?

kvarsh commented 1 year ago

In-processing algorithms are tied to their models. Specific other classifiers cannot be swapped in.

pradeepdev-1995 commented 1 year ago

So applying in processing algorithm to mitigate bias in an existing model(SVM, logistic regression,MLP...etc) is not possible in AIF360? @kvarsh

kvarsh commented 1 year ago

The whole idea of in-processing algorithms is that they are models themselves. For example the prejudice remover is a logistic regression regularized by a fairness term. https://doi.org/10.1007/978-3-642-33486-3_3 https://aif360.readthedocs.io/en/v0.4.0/modules/generated/aif360.algorithms.inprocessing.PrejudiceRemover.html

This is not related to AIF360, but just the nature of in-processing.