Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Possible solution for Sparse network see also #67 #136

Open
mkostrun opened this issue Jan 30, 2023 · 1 comment
Open

Possible solution for Sparse network see also #67 #136

mkostrun opened this issue Jan 30, 2023 · 1 comment

Comments

@mkostrun
Copy link

From my previous work on fann-like libraries it would be beneficial to associate with weights an array "is_mutable". In case if a certain weight is set to, say, zero, and its is_mutable set to zero, as well, then during optimization this weight stays immutable, that is, fixed to its initial value.
I did a fork of fann for rlabplus, that implements this mechanics, and if you want we could merge it with current fann.

@mkostrun
Copy link
Author

I am following up with my past self:
Other benefit of having immutable flag for each weight is that then the cascade/shortcut network as needing specialized treatment, become unnecessary.
By adding to each layer past the input one a number of linear nodes, where the number of nodes is equal to the sum of the outputs of all previous layers; then setting the weights of these additional nodes to [1,0,0,0] and so forth; and then making them immutable; one creates an equivalent structure to the shortcut network.
In other words, the shortcut network is just a special case of a general feed-forward network, where certain weights are fixed, so it doesn't need special treatment.
However, for this to work one needs immutable flag for the weight and bias arrays, as proposed in my first post.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

1 participant