Superpolynomial Lower Bounds on Learning One-Layer Neural Nets with Gradient Descent

ICML 2020