A Simple Yet Effective Strategy to Robustify the Meta Learning Paradigm