Graph Neural Networks with polynomial activations have limited expressivity.
Sammy KhalifePublished in: CoRR (2023)
Keyphrases
- neural network
- pattern recognition
- graph theoretic
- graph theory
- random walk
- artificial neural networks
- back propagation
- graph representation
- connected components
- graph databases
- graph mining
- genetic algorithm
- graph structure
- graph matching
- fuzzy logic
- multilayer perceptron
- multi layer
- fuzzy systems
- network architecture
- neural network model
- activation function
- web graph
- fault diagnosis
- rule extraction
- low order