Doctoral Dissertation Defense: Francesca McFadden

As industries adopt workflows based on recommendations from machine-learned models, there is an increased need to ensure appropriate trust in the application of the predictions for end-user application or downstream system use. Model competence methods extend current application of model reported confidence to ensure that models are being trusted on input data they have appropriate prediction capability and corresponding training (data) for. We apply unsupervised learning, namely clustering methods, to enhance current approaches to model competence estimation. We review clustering methods which may be used for out-of-distribution indication and distance metrics that we consider in their employment, including the Mahalanobis distance.

Three applications are described and aim to be documented in appropriate conference presentations and/or proceedings. One of the topics was presented at the Society of Industrial and Applied Mathematics (SIAM) Mathematics of Data Science (MDS) 2024 conference in October 2024 and a second topic was presented in April 2025 at the DATAWorks conference in Alexandria, VA.