Learning ReLU networks to high uniform accuracy is intractable

Open in new window