I have some background in metaheuristics based optimization but not so much in machine learning. I get the feeling that metaheuristic methods are not commonly employed in machine learning. Is this primarily due to the objective function evaluation cost?