. "where b is a N dim vector, M is a N by N matrix (no assumption for sparseness), and g is some nonlinear function." . . .