Neural Optimal Design of Experiment for Inverse Problems