We discuss how to decompose the excess risk into the estimation, approximation and optimization error.
