Toward Moderate Overparameterization: Global Convergence Guarantees for Training Shallow Neural Networks | IEEE Journals & Magazine | IEEE Xplore