2 Fairness and mitigating bias

 

Chapter 8 from Interpretable AI by Ajay Thampi

This chapter covers

  • Identifying sources of bias in datasets
  • Validating if machine learning models are fair using various fairness notions
  • Applying interpretability techniques to identify the source of discrimination in machine learning models
  • Mitigating bias using pre-processing techniques
  • Documenting datasets using datasheets to improve transparency and accountability, and to ensure compliance with regulation

8.1 Adult Income Prediction

8.1.1 Exploratory Data Analysis

8.1.2 Prediction Model

8.2 Fairness Notions

8.2.1 Demographic Parity

8.2.2 Equality of Opportunity and Odds

8.2.3 Other Notions of Fairness

8.3 Interpretability and Fairness

8.3.1 Discrimination via Input Features

8.3.2 Discrimination via Representation

8.4 Mitigating Bias

8.4.1 Fairness through Unawareness

8.4.2 Correcting Label Bias through Re-weighting

8.5 Datasheets for Datasets

8.6 Summary