On November 19th, the second seminar of the CoMeEcon series will take place in Amsterdam.
Our speaker, Martijn Gösgens from the CWI, 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
(+ accessible online, more info coming soon)
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.