Saved in:
Bibliographic Details
Main Authors: Denton, Remi, Hutchinson, Ben, Mitchell, Margaret, Gebru, Timnit, Zaldivar, Andrew
Format: Preprint
Published: 2019
Subjects:
Online Access:https://arxiv.org/abs/1906.06439
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866911794733252608
author Denton, Remi
Hutchinson, Ben
Mitchell, Margaret
Gebru, Timnit
Zaldivar, Andrew
author_facet Denton, Remi
Hutchinson, Ben
Mitchell, Margaret
Gebru, Timnit
Zaldivar, Andrew
contents Facial analysis models are increasingly used in applications that have serious impacts on people's lives, ranging from authentication to surveillance tracking. It is therefore critical to develop techniques that can reveal unintended biases in facial classifiers to help guide the ethical use of facial analysis technology. This work proposes a framework called \textit{image counterfactual sensitivity analysis}, which we explore as a proof-of-concept in analyzing a smiling attribute classifier trained on faces of celebrities. The framework utilizes counterfactuals to examine how a classifier's prediction changes if a face characteristic slightly changes. We leverage recent advances in generative adversarial networks to build a realistic generative model of face images that affords controlled manipulation of specific image characteristics. We then introduce a set of metrics that measure the effect of manipulating a specific property on the output of the trained classifier. Empirically, we find several different factors of variation that affect the predictions of the smiling classifier. This proof-of-concept demonstrates potential ways generative models can be leveraged for fine-grained analysis of bias and fairness.
format Preprint
id arxiv_https___arxiv_org_abs_1906_06439
institution arXiv
publishDate 2019
record_format arxiv
spellingShingle Image Counterfactual Sensitivity Analysis for Detecting Unintended Bias
Denton, Remi
Hutchinson, Ben
Mitchell, Margaret
Gebru, Timnit
Zaldivar, Andrew
Computer Vision and Pattern Recognition
Machine Learning
Facial analysis models are increasingly used in applications that have serious impacts on people's lives, ranging from authentication to surveillance tracking. It is therefore critical to develop techniques that can reveal unintended biases in facial classifiers to help guide the ethical use of facial analysis technology. This work proposes a framework called \textit{image counterfactual sensitivity analysis}, which we explore as a proof-of-concept in analyzing a smiling attribute classifier trained on faces of celebrities. The framework utilizes counterfactuals to examine how a classifier's prediction changes if a face characteristic slightly changes. We leverage recent advances in generative adversarial networks to build a realistic generative model of face images that affords controlled manipulation of specific image characteristics. We then introduce a set of metrics that measure the effect of manipulating a specific property on the output of the trained classifier. Empirically, we find several different factors of variation that affect the predictions of the smiling classifier. This proof-of-concept demonstrates potential ways generative models can be leveraged for fine-grained analysis of bias and fairness.
title Image Counterfactual Sensitivity Analysis for Detecting Unintended Bias
topic Computer Vision and Pattern Recognition
Machine Learning
url https://arxiv.org/abs/1906.06439