-
Notifications
You must be signed in to change notification settings - Fork 3
Implement Sparse Coding #2
Comments
Rethinking this a bit, I don't think this is correct:
If you consider the input |
I think there are two separate problems to consider, the Dictionary Learning problem (learning |
Sparse Coding is a framework of finding basis vectors and weights to decompose an input vector into the linear combination of a subset of basis vectors. The framework is very similar to Empirical Risk Minimization in that there's a loss and a penalty, except that the penalty is on the output, not the parameters. I believe sparse coding refers specifically to a linear model, but of course any combinations of transformation/loss/penalty could be used... we just need to add a penalty on the output. Should we just go ahead and add this to
RegularizedObjective
? It can default toNoPenalty
.ref:
http://ufldl.stanford.edu/wiki/index.php/Sparse_Coding
http://gabgoh.github.io/SARG/
cc: @gabgoh
The text was updated successfully, but these errors were encountered: