Does a sparse ReLU network training problem always admit an optimum?

Re-assign community