Mittagsseminar (by J. Lengler, K. Bringmann, B. Gärtner, M. Hoffmann, R. Kyng, D.Steurer, V. Traub)

Date and Time: Tuesday, March 04, 2025, 12:15 pm

Duration: 30 minutes

Location: CAB G51

Speaker: Maxim Mikhaylov

Topology of decision boundaries

How do machine learning decision boundaries evolve during training, and what can their topology reveal about model performance? In this talk, we explore the use of persistent homology to analyze decision boundaries of classifiers, tracking their topological changes over the course of training. We introduce a multiclass Labeled Vietoris-Rips complex, a more accurate and efficient approach for capturing the decision boundary. Our results show that topological metrics strongly correlate with model accuracy, even in cases of overfitting. No prior knowledge of machine learning or topological data analysis is expected - relevant concepts will be introduced.

Upcoming talks     |     All previous talks     |     Talks by speaker     |     Upcoming talks in iCal format (beta version!)

Previous talks by year:   2026  2025  2024  2023  2022  2021  2020  2019  2018  2017  2016  2015  2014  2013  2012  2011  2010  2009  2008  2007  2006  2005  2004  2003  2002  2001  2000  1999  1998  1997  1996  

Information for students and suggested topics for student talks

Automatic MiSe System Software Version 1.4803M   |   admin login