Confidence Intervals for Error Rates in 1:1 Matching Tasks: Critical Statistical Analysis and Recommendations

Riccardo Fogliato,Pratik Patil,Pietro Perona
DOI: https://doi.org/10.1007/s11263-024-02078-8
IF: 13.369
2024-06-13
International Journal of Computer Vision
Abstract:Matching algorithms predict relationships between items in a collection. For example, in 1:1 face verification, a matching algorithm predicts whether two face images depict the same person. Accurately assessing the uncertainty of the error rates of such algorithms can be challenging when test data are dependent and error rates are low, two aspects that have been often overlooked in the literature.In this work, we review methods for constructing confidence intervals for error rates in 1:1 matching tasks. We derive and examine the statistical properties of these methods, demonstrating how coverage and interval width vary with sample size, error rates, and degree of data dependence with experiments on synthetic and real-world datasets. Based on our findings, we provide recommendations for best practices for constructing confidence intervals for error rates in 1:1 matching tasks.l
computer science, artificial intelligence
What problem does this paper attempt to address?