Graduate Thesis Or Dissertation
 

Individual Fairness Optimization in Machine Learning with Minimax Loss and an Abstain Option

Pubblico Deposited

Contenuto scaricabile

Scarica il pdf
https://ir.library.oregonstate.edu/concern/graduate_thesis_or_dissertations/db78tm144

Descriptions

Attribute NameValues
Creator
Abstract
  • In this thesis, a new learning algorithm is introduced that is targeted towards individual fairness. In order to be individually fair, mispredictions need to be avoided as each such prediction means the learning algorithm was unfair towards some individual. Therefore, achieving individual fairness implies having a perfect classifier, which is hard to achieve in practice. The approach adopted here attempts to make it easier to achieve such a classifier by introducing a trade-off: augmenting classification with the rejection option. Essentially, making fewer mispredictions to improve fairness at the expense of lower coverage, i.e. making predictions for only a subset of samples. Decision to reject or make a prediction for each sample is driven by uncertainty estimates. As such, having well-calibrated uncertainty estimates is paramount to the success of this approach. To that end and due to the success of Stein Variational Gradient Descent (SVGD) in recent years, this method was incorporated into our approach. Furthermore, a growing body of research in recent years have emphasized the need to directly involve end-users and their perception of fairness in the design of fair learning algorithms. To facilitate such human-in-the-loop approach, a web-based user-interface was implemented for the proposed algorithm. Through this user-interface, end-users can explore the data, adjust various parameters and view the predictions made by the algorithm.
License
Resource Type
Date Issued
Degree Level
Degree Name
Degree Field
Degree Grantor
Commencement Year
Advisor
Committee Member
Academic Affiliation
Dichiarazione dei diritti
Publisher
Peer Reviewed
Language

Le relazioni

Parents:

This work has no parents.

In Collection:

Elementi

Thumbnail Titolo Data caricata Visibilità Azioni
file details BehnoudfarAli2022.pdf 2022-09-14 Pubblico Scaricare