SSLaP/DyS
Seminar of Statistical Learning and Probability/Dynamical Systems
University of Leipzig, ScaDS.AI Institute,
Max Planck Institute for Mathematics in the Sciences &
University of Pisa, Department of Mathematics
Upcoming seminar
April 8th 2024, 16:00 CET
Maximilian Wiesmann (Max Planck Institute for Mathematics in the Sciences)
Title: Adversarial Regularization Regimes in Classification Tasks
Abstract: In this talk we demonstrate the possibility of a trend reversal in binary classification tasks between the dataset and a classification score obtained from a trained model. This trend reversal occurs for certain choices of the regularization parameter for model training, namely, if the parameter is contained in what we call the adversarial regularization regime. For ridge regression, we give necessary and sufficient algebraic conditions on the dataset for the existence of an adversarial regularization regime. Moreover, our results provide a data science practitioner with a hands-on tool to avoid hyperparameter choices suffering from trend reversal. We furthermore present numerical results on adversarial regularization regimes for logistic regression. Finally, we draw connections to datasets exhibiting Simpson’s paradox, providing a natural source of adversarial datasets.
In-person seminars take place in Paulinum P701, Leipzig University or in Aula Riunioni, Dipartimento di Matematica, University of Pisa and are also broadcasted on Zoom, join at the following link.
Andrea Agazzi
University of Pisa
Nono Saha Cyrille
Leipzig University (ScaDS.AI)
Alexander Kreiss
Leipzig University
Sayan Mukherjee
Leipzig University and Max-Planck-Institut for Mathematics in the Sciences
Katerina Papagiannouli
University of Pisa and Max-Planck-Institut for Mathematics in the Sciences