You may try fsolve(). However, due to the complexity of a neural network, it is unlikely that it will be able to find a solution.
fsolve(@(x) net(x)-out_vec, rand(8,1));
You may also try optimization based methods
fmincon(@(x) norm(net(x)-[1.5 -0.5]), rand(3,1))
ga(@(x) norm(net(x(:))-[1.5 -0.5]), 3)