On Equivalent Optimization of Machine Learning Methods