Extended ResNet

The success of ResNet is mainly ascribed to the fact that it is very easy for deeper networks to learn the identity function hence there’s little risk when the network becomes too deep. I was wondering why don’t we take that even further and say for any given block $n$ comprised of a sequence of operations such as Conv2D, BatchNorm, and Relu, the block would be connected to all the previous blocks labeled 1,…,$n-1$ as opposed to just $n-2$ for example.