Sign in

Pascal Welke

@pascalwelke.bsky.social
145 followers 78 following 15 posts

Graph Machine Learning and Graph Mining Assistant Professor (Lecturer) in Data Science Lancaster University Leipzig pwelke.de

PostsRepliesMedia
Pascal Welke @pascalwelke.bsky.social · 11/06/2025
Attended my first @netsciconf.bsky.social last week and it was amazing! A highlight was the HONAI satellite that brought together the network science and machine learning communities. ...and who needs NeurIPS mugs if one can have NetSci toilet paper?
010
Pascal Welke @pascalwelke.bsky.social · 15/12/2024
Is Expressivity Essential for the Predictive Performance of Graph Neural Networks? Spoiler alert: No. Check out our poster at the Sci4DL workshop, today at 4.30pm, West Meeting Room 205-207 Paper: pwelke.de/publications... Poster: pwelke.de/publications...
TL;DR More expressive GNNs outperform less expressive GNNs not due to expressivity. 

Pdf Version of the poster, as well as the paper is available at https://pwelke.de
0100
Pascal Welke @pascalwelke.bsky.social · 13/12/2024
Today at NeurIPS: Weisfeiler and Leman go Loopy: A New Hierarchy for Graph Representational Learning Cycles are important for predictive tasks on chemical molecules. We allow message passing along neighboring paths. Our architecture can subgraph-count cycles and homomorphism-count cactus graphs.
Visual depiction of r-lGIN: During preprocessing, we calculate the path neighborhoods Nr (v) for each node v in the graph G. Paths of varying lengths are processed separately using simple GINs, and their embeddings are pooled to obtain the final graph embedding. The forward complexity scales linearly with the sizes of Nr (v), enabling efficient computation on sparse graphs.
161