2021
DOI: 10.1007/978-3-030-88853-4_2
|View full text |Cite
|
Sign up to set email alerts
|

A Logical Characterization of Constant-Depth Circuits over the Reals

Abstract: We characterize the computational power of neural networks that follow the graph neural network (GNN) architecture, not restricted to aggregatecombine GNNs or other particular types. We establish an exact correspondence between the expressivity of GNNs using diverse activation functions and arithmetic circuits over real numbers. In our results the activation function of the network becomes a gate type in the circuit. Our result holds for families of constant depth circuits and networks, both uniformly and non-… Show more

Help me understand this report

Search citation statements

Order By: Relevance

Paper Sections

Select...
3
1
1

Citation Types

0
0
0

Year Published

2024
2024
2024
2024

Publication Types

Select...
1

Relationship

0
1

Authors

Journals

citations
Cited by 1 publication
(5 citation statements)
references
References 21 publications
(13 reference statements)
0
0
0
Order By: Relevance
“…For future work, it would be interesting to investigate the logical characterizations made in this paper in the uniform setting. We know that for the real numbers, the characterization AC 0 R = FO R [Arb R ] + SUM R + PROD R holds both non-uniformly and for uniformity criteria given by polynomial time computable circuits (P R uniform), logarithmic time computable circuits (LT R uniform) and first-order definable circuits (FO R uniform) (Barlag and Vollmer, 2021). We believe that the results we presented here hold in analogous uniform settings as well, though this would need to be further examined.…”
Section: Discussionmentioning
confidence: 99%
See 4 more Smart Citations
“…For future work, it would be interesting to investigate the logical characterizations made in this paper in the uniform setting. We know that for the real numbers, the characterization AC 0 R = FO R [Arb R ] + SUM R + PROD R holds both non-uniformly and for uniformity criteria given by polynomial time computable circuits (P R uniform), logarithmic time computable circuits (LT R uniform) and first-order definable circuits (FO R uniform) (Barlag and Vollmer, 2021). We believe that the results we presented here hold in analogous uniform settings as well, though this would need to be further examined.…”
Section: Discussionmentioning
confidence: 99%
“…Proof. The proof for this theorem works similarly to the construction for FO R and Vollmer, 2021), since this construction does not make use of any special properties of the real numbers.…”
Section: Acmentioning
confidence: 99%
See 3 more Smart Citations