Accurate neural networks can be found just by pruning a randomly initialized overparameterized model. leaving out the need for any weight optimization. The resulting subnetworks are small. sparse. https://thebrickes.shop/product-category/pet-combs/