I am a student. I am currently looking into graph neural networks (GNNs). My domain is electrical power systems. In electrical power systems, it is extremely important that we get an accurate desired output numerical value of electrical data from a neural network.
1) I have a basic question. Consider an electrical grid network of nodes. I am trying to learn this electrical grid network data using Graph Neural Network (GNN). Every node of a GNN accumulates data from neighboring nodes, then processes it by a few steps of an algorithm, and passes it to the next layer. Finally, data is passed through a non-linearity and then to the output layer of the GNN.
But, if I feed electrical data to the above process, the original value of data at every node gets manipulated by several processing operations, and especially after passing the manipulated data through a non-linearity at the final stage, the output is obtained only in the form of 1's and 0s. Hence, the original electrical data value at every node is totally lost. On the contrary, I am expecting an output of an "accurate" value of electrical data similar to original value electrical data at every node of the network.
How to address the above problem? Please explain systematically if possible. This is a genuine basic question.
2) Also, does anyone have a clue, why Graph Neural Networks (GNNs) have not been introduced yet as a toolbox or in general in Matlab?
Help and opinion on above questions would be greatly appreciated.