Skip to content

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

29 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

KAN

Experiments with learnable KAN / univariate functions.

The actual main code for training networks are in tests across the various packages.

Status: Probably not going to work on this anymore as Bezier KANs seem to not train without residual functions.

Ideas backlog

  • Some loss penality for non-convexity?
  • Squimble-lock: dynamically reduce learning rate if gradient at each step oscillates
    • What if we could characterize the oscillation + negate it?
  • What if we adjust the weights beyond the spline it contributes to?
  • What if we try and smooth out the spline every N steps?

About

No description, website, or topics provided.

Resources

Stars

Watchers

Forks

Releases

Packages

Used by

Contributors

Languages