ResNet Question: How can it skip "some Layers" of the NN given that each Layer needs the output of the previous Layer?

[ralsenger]

H​i all,

I​ didn’t understand the skip approach from ResNets.

G​iven that:

If I skip the calculation of Layer [l+1], then a[l+1] isn’t calculated, but a[l+2] depends on this value.

What did I miss in this approach?

Thanks in advance.

[Dyxuki]

Hi,
no you do not skip layers, but rather you can understand the formula as adding an extra term which come from 2 layers before.

For example:
a[1] = g( W[1]a[0] + b[1])
a[2] = g(W[2]a[1] + b[2] + a[0])
a[3] = g(W[3]a[2] + b[3] + a[1])
a[4] = g(W[4]a[3] + b[4] + a[2])
etc.
all the bold terms are the “residues”, namely extras, if you remove them then you have “standard FClayers”