Off-campus University of Richmond users: To download campus access theses, please use the following link to log in to our proxy server with your university username and password.
Date of Award
4-29-2022
Document Type
Restricted Thesis: Campus only access
Degree Name
Bachelor of Science
Department
Computer Science
First Advisor
Douglas Szajda
Abstract
Machine learning systems are being developed to solve all sorts of problems. Some of the applications include determining whether to give someone a loan, the likelihood of recidivism, which areas in a community are more likely to have crime, whether computer code is malicious or benign, or whether an autonomous vehicle is approaching a stop sign. All these applications can have a significant impact on people’s lives, yet we don’t fully understand the "reasoning" behind these models. Users of such models need to know whether they contain biases and whether they are basing decisions on reasonable principles. These needs have given rise to explanation methods, which attempt to expose the reasoning behind specific machine learning model decisions. However, through our research, we have found that explanation methods are not as reliable as we would like them to be, hence our attempt to create provably robust explanation methods.
Recommended Citation
Shannon, André, "Developing Provably Robust Explanation Methods for Image Classifiers" (2022). Honors Theses. 1656.
https://scholarship.richmond.edu/honors-theses/1656