Theory II: Landscape of the Empirical Risk in Deep Learning