Appendix of Optimization and Generalization of Shallow Neural Networks with Quadratic Activation Functions