symbolic regression of KAN for formular systems

[splendid249]

when I used the KAN for symbolic regression, I got some weird activation functions which is shown in the following picture. so is there any method to make the formular smoother?

[wing2310]

I have the same question with you. so is the problem solved? if yes, could you share your approach? thank you.