The workflow for dlnetwork and trainnet would be something like the following:
image = randi(255,[3,3,4]);
% create network
net = [
imageInputLayer(size(image),Name="data",Normalization="none")
additionLayer(2,Name="add")];
% create as uninitialized so you can hook up the 2nd input to additionLayer before initializing
net = dlnetwork(net,Initialize = false);
net = connectLayers(net,"data","add/in2");
% initialize
net = initialize(net);
% train
opts = trainingOptions("adam");
% regressionLayer is not used, to specify loss you can use "mse" in trainnet
trainnet(image,image,net,"mse",opts);
