--

Simply explained with some good humorous analogy. It follows by this logic that the output of a ResNet block (indicated by the green by-pass) can be at least equal to the input to that block even if weights and biases decay to zero. When we are using ReLU activation units.

--

--

Kizito Nyuytiymbiy
Kizito Nyuytiymbiy

Written by Kizito Nyuytiymbiy

Transformational Speaker | Effective communication/Public Speaking Trainer/Coach | kizitonyuytiymbiy.com | https://twitter.com/Kizito

Responses (1)