Provable Generalization of Overparameterized Meta-learning Trained with SGD

Neural Information Processing Systems 

Global convergence and induced kernels of gradient-based meta-learning with neural nets.