Record link: 
http://hdl.handle.net/20.500.12708/216592
-
Title: 
Is Expressivity Essential for the Predictive Performance of Graph Neural Networks?
en
Citation: 
Jogl, F., Welke, P., & Gärtner, T. (2024). Is Expressivity Essential for the Predictive Performance of Graph Neural Networks? In NeurIPS 2024 Workshop on Scientific Methods for Understanding Deep Learning. NeurIPS 2024 The Thirty-Eighth Annual Conference on Neural Information Processing Systems, Vancouver, Canada.
-
Publication Type: 
Inproceedings - Full-Paper Contribution
en
Language: 
English
-
Authors: 
Jogl, Fabian 
Welke, Pascal 
Gärtner, Thomas 
-
Organisational Unit: 
E192-02 - Forschungsbereich Databases and Artificial Intelligence 
E194-06 - Forschungsbereich Machine Learning 
E056-10 - Fachbereich SecInt-Secure and Intelligent Human-Centric Digital Technologies 
E056-23 - Fachbereich Innovative Combinations and Applications of AI and ML (iCAIML) 
E056-26 - Fachbereich Automated Reasoning 
-
Published in: 
NeurIPS 2024 Workshop on Scientific Methods for Understanding Deep Learning 
-
Date (published): 
2024
-
Event name: 
NeurIPS 2024 The Thirty-Eighth Annual Conference on Neural Information Processing Systems 
en
Event date: 
10-Dec-2024 - 15-Dec-2024
-
Event place: 
Vancouver, Canada
-
Number of Pages: 
10
-
Peer reviewed: 
Yes
-
Keywords: 
Graph Neural Networks; Expressivity; Machine Learning
en
Abstract: 
Motivated by the large amount of research on the expressivity of GNNs, we study the impact of expressivity on the predictive performance of GNNs. By performing knowledge distillation from highly expressive teacher GNNs to less expressive student GNNs, we demonstrate that knowledge distillation reduces the predictive performance gap between teachers and students significantly. As knowledge distillation does not increase the expressivity of the student GNN, it follows that most of this gap in predictive performance cannot be due to expressivity.
en
Project title: 
Structured Data Learning with Generalized Similarities: ICT22-059 (WWTF Wiener Wissenschafts-, Forschu und Technologiefonds) 
-
Additional information: 
conference: https://neurips.cc/Conferences/2024 -- "proceedings": https://openreview.net/forum?id=5v7hpSy3Ir paper: https://openreview.net/attachment?id=5v7hpSy3Ir&name=pdf
-
Research Areas: 
Information Systems Engineering: 100%
-
Science Branch: 
1020 - Informatik: 100%
-
Appears in Collections:
Conference Paper

Show full item record

Page view(s)

12
checked on Jul 2, 2025

Google ScholarTM

Check