GNN Class¶

Graph neural network class supporting GCN and GINE architectures with node and edge feature uncertainty via GraphStar sets.

Constructors¶

gnn = GNN(layers)                                      % Layers only
gnn = GNN(layers, A_norm)                              % GCN mode
gnn = GNN(layers, A_norm, adj_list, E)                 % GINE mode
gnn = GNN(layers, A_norm, adj_list, E, edge_weights)   % GINE with edge weights

layers – cell array of GCNLayer, GINELayer, ReluLayer, etc.
A_norm – normalized adjacency matrix (N x N)
adj_list – edge list [src, dst] (M x 2)
E – edge feature matrix (M x d_e) or EdgeGraphStar
edge_weights – optional edge weight vector

Properties¶

Property	Type	Description
`Layers`	cell array	GCN/GINE and activation layers
`A_norm`	matrix	Normalized adjacency matrix
`adj_list`	matrix	Edge list (M x 2)
`E`	matrix	Edge feature matrix
`edge_weights`	vector	Optional edge weights

Methods¶

Method	Description
`Y = gnn.evaluate(X)`	Evaluate GNN on node feature matrix X (N x F_in). Returns output node features.
`outputSet = gnn.reach(inputSet, reachOptions)`	Compute reachable set. Input/output: GraphStar.
`gnn.setGraph(A_norm, adj_list, E)`	Update graph structure (allows weight reuse across different graphs).

Example¶

layers = {GCNLayer(W1, b1), ReluLayer(), GCNLayer(W2, b2)};
gnn = GNN(layers, A_norm);

Y = gnn.evaluate(X_test);

GS = GraphStar(X, -eps_matrix, eps_matrix);
reachOptions.reachMethod = 'approx-star';
output_sets = gnn.reach(GS, reachOptions);