Approximate Computation and Implicit Regularization for Very Large-scale Data Analysis