The hypothesis that dense networks contain small subnetworks which, trained in isolation from the same initial weights, can reach comparable accuracy.
Continue to AI University →