Session 9: Iterative Estimation of Coefficients for Generalized Linear Models on Pairwise Scores
Presentation Type
Invited
Student
No
Track
Forensic Statistics
Abstract
Pairwise scores are commonly used in biometric settings for identification. These are often black-box scores, where their inner workings are difficult to understand. Fuglsby (2023) introduced methods to compare different biometric algorithms for the purpose of explainability. One such method is the Generalized Linear Model (GLM). This model requires the inverse of the covariance matrix on the dependent variable conditional on the independent variable, which is difficult to estimate from pairwise scores. This talk will introduce an algorithm for estimating the parameters necessary for estimating the covariance matrix.
Start Date
2-6-2024 2:30 PM
End Date
2-6-2024 3:30 PM
Session 9: Iterative Estimation of Coefficients for Generalized Linear Models on Pairwise Scores
Pairwise scores are commonly used in biometric settings for identification. These are often black-box scores, where their inner workings are difficult to understand. Fuglsby (2023) introduced methods to compare different biometric algorithms for the purpose of explainability. One such method is the Generalized Linear Model (GLM). This model requires the inverse of the covariance matrix on the dependent variable conditional on the independent variable, which is difficult to estimate from pairwise scores. This talk will introduce an algorithm for estimating the parameters necessary for estimating the covariance matrix.