Convex losses and their applications to SVM, SVR, and Shallow Neural Networks