Towards Sample-efficient Overparameterized Meta-learning