Jan 26, 2023 · In this paper, we study GNNs' generalization ability through the lens of Vapnik--Chervonenkis (VC) dimension theory in two settings, focusing on graph-level ...
Here, the 1-WL is a well-studied heuristic for the graph isomorphism problem, which iteratively colors or partitions a graph's vertex set. While this connection ...
We show that GNNs' VC dimension depends tightly on the number of equivalence classes computed by the 1-WL over a set of graphs; see Propositions 1 and 2.
Jul 23, 2023 · In this paper, we study GNNs' generalization ability through the lens of Vapnik-Chervonenkis (VC) dimension theory in two settings, focusing on ...
WL meet VC ±. We show a tight connection between GNNs' expressivity and their generalization ability. 7. Page 9. WL meet VC ±. Uniform? Bitlength ≤ b? = b. 1-WL ...
In this paper, we study GNNs' generalization ability through the lens of Vapnik--Chervonenkis (VC) dimension theory in two settings, focusing on graph-level ...
Algorithms, Dynamics, and Information Flow in Networks DFG Research Unit. ≡ Open menu. Home · Projects · People · Publications · Events · Media · Contact.
WL meet VC. Publication. WL meet VC. Morris, Christopher (Corresponding author); Geerts, Floris; Tönshoff, Jan Martin; Grohe, Martin. ML Research Press (2023)
People also ask
What is the VC-dimension of a neural network?
What is the VC-dimension of a graph?
The aim of our work is to extend this analysis on the VC dimension of GNNs to other commonly used activation functions, such as the sigmoid and hyperbolic ...
Jul 13, 2024 · Each week, we want to highlight LoG papers! ✨ Check out "WL meet VC" by Christopher Morris, Floris Geerts, Jan To nshoff, and Martin Grohe!