Performance of Kolmogorov-Arnold network

[A P]

Hello everyone,

Some people noticed new paper of Max Tegmark about their way of building Kolmogorov-Arnold network.  They use Broyden algorithm, which is slow.

I also work with this model since 2020 (they have reference on me in their paper).  My approach is Kaczmarz method. 

I try to compare my implementation with others, it can be found on 

http://openkan.org/

My C++ version train 10 000 records (5 features) for 0.1 second. It is not a big deal.

It is only near MLP. 

Who knows other C++ implementations of KAN?

Thanks.