I use KAN KAN KAN KAN KAN KAN KAN KAN KAN KAN KAN KAN KAN
Honestly i dont know what to write here
Activation Function learns rather than weight i suppose, replacing every weight with spline (Spine much better)
Proved to be better performance than MLP (Multilayer perceptron)
Still some doubt because it was tested on small dataset
Pros:
- Easier to iterpret
- Smaller storage needed
Cons:
- slow af evethough usig the same amout of parameter to MLP
Check out the paper: https://arxiv.org/abs/2404.19756
Very gud paper but i can't understand the math and I can't read many sentences (me dumb)