Detecting Racial Bias

I will first introduce the general premise of statistical racial bias with some specific examples fake examples from:

• education - not getting accepted to a college
• law enforcement - racial profiling
• finance and banking - not getting approved for a loan

A review of model Interpretability

Here I will introduce model interpretability and discuss specific metrics used. We will work through:

• Linear Regression
• log-log features
• linear-log features
• log-linear features
• non-linear features
• T-test
• F-test
• MSE
• R^2
• Tree-Models
• Feature Importance
• MSE
• Tree Interpretter
• Neural Networks
• learning a tree based model from the results
• (LRP) Layer-wise Relevance Propagation
• DeepLift
• DeepExplain
• General Tools
• LIME
• SHAP

Each tool will be used to look for racial bias in our faked data.

Introduction of Algorithmic Fairness Testing tools

Here I will walk through some of the tools used to test for racial bias and apply them to the datasets:

• Fairtest - https://github.com/columbia/fairtest
• FairML - https://github.com/adebayoj/fairml