Question about weights #631
Closed
rafaelblevin821
started this conversation in
General
Replies: 2 comments
-
I figured it out for anyone wanting to know how the weights are calculated. The 79 was coming from the value that 'norm' is set to. So 'Norm' = (Sum of all weight connections) / (number of neurons) |
Beta Was this translation helpful? Give feedback.
0 replies
-
While there may not be a specified reason for this normalization, it's possible that the "normal" ratio between black and white pixels in the MNIST dataset has been taken into consideration. This is just my speculation though, and perhaps someone else has a more informed explanation. |
Beta Was this translation helpful? Give feedback.
0 replies
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
-
Could anyone tell me why:
(Sum of all weight connections) / (number of neurons) = 79 ???
eg. Summatation of all input connection weights = (55300) / n_neurons (700) = 79
eg. Summatation of all input connection weights = (63200) / n_neurons (800) = 79
... and so on, for every value of n_neurons, the summation of the weights will always equate to: n_neurons * 79.
I am using the Diehl and Cook supervised mnist model.
Is there a piece of code that locks the weights to a specific value per number of neurons?
Beta Was this translation helpful? Give feedback.
All reactions