Calculating a boundary for the significance from the equal-error rate

摘要

Given a common dataset, two methods operating on that dataset and reported equal-error rate (EER) for each method, then we can estimate whether the two methods differ significantly at the threshold leading to the EER. This enables the calculation of a boundary on the significance for methods where the significance was not reported in the original paper or to compare new methods to older ones by evaluating them on the same dataset.