On November 19th, the second seminar of the CoMeEcon series will take place. Nuria and Kayané from the Big Statistics group (EDS department) of the Amsterdam UMC would like to invite you to this winter episode of our CoMeEcon seminar series!

Our speaker, Martijn Gösgens, from CWI Amsterdam, will tell us about “Validating validation measures for clustering and classification tasks”. Drinks will follow the presentation.

Date: Wednesday the 19th of November 2025
Time: 14.00 – 17.00
Location: Vrije Universiteit Amsterdam (and accessible online)

Abstract
In machine learning, many algorithms can be used for the same task, and benchmark experiments are often used to compare them. These experiments apply algorithms to datasets with known ground truth and then evaluate performance by comparing predictions with the ground truth. This seminar will focus on that last step: how should we measure the quality of an algorithm’s output? We will consider the classical problems of classification and clustering, for which many validation measures exist (e.g., the F1 score for classification or NMI for clustering). In practice, researchers often rely on the most commonly used measures, but many of these have severe biases. In this seminar, I will highlight these issues and present alternative measures that address them.

COMEECON is a seminar on Computational statistics, Methodology and Econometrics aimed at early career statisticians. It is part of the Mathematical Statistics section of the VVSOR.