Understanding and reasoning fairness in machine learning pipelines

By: Sumon Biswas

Download Paper

Abstract

This dissertation studies fairness in machine learning pipelines. It introduces a method for reasoning about the composition of group fairness across the stages of a pipeline, a large-scale empirical study of unfairness in open-source machine learning models, and Fairify, a technique for verifying individual fairness in neural networks. Together these contributions help developers see where unfairness enters a pipeline and reason about it rigorously rather than only testing the final model.

ACM Reference

Biswas, S. 2022. Understanding and reasoning fairness in machine learning pipelines. Iowa State University.

BibTeX Reference

@phdthesis{Biswas2022,
  title = {Understanding and reasoning fairness in machine learning pipelines},
  author = {Biswas, Sumon},
  year = {2022},
  school = {Iowa State University},
  abstract = {
    This dissertation studies fairness in machine learning pipelines. It introduces a
    method for reasoning about the composition of group fairness across the stages of a
    pipeline, a large-scale empirical study of unfairness in open-source machine learning
    models, and Fairify, a technique for verifying individual fairness in neural networks.
    Together these contributions help developers see where unfairness enters a pipeline and
    reason about it rigorously rather than only testing the final model.
  }
}