I am faced with a problem which posed the following question.
Suppose a set of graphs, where each graph can be composed of, at least, two nodes, and at max, five nodes. Every node has a fixed size node feature, and connectivity can vary in any way in each of the graphs.
Suppose that we label the graphs in the following way: If a graph is composed of four nodes and every node is connected to each other, label = 1. Otherwise, label = 0.
Which type of graph neural network can “learn” this labeling trivially?
P.S. : This is a simplified problem, as in reality I must take node features into account.
A simplified approach can be to described graph connectivity as a feature to a NN, but I specifically wanted to know if GNN can handle this type of problem.