Hi all,
Just wondering if someone has already made attempt at implementing a Kolmogorov-Arnold layer.

See [2404.19756v2] KAN: Kolmogorov-Arnold Networks

@kgoderis if you want to do anything custom you’d usually want to do that with the samediff api and implement it as a SameDiffLayer if you want to use the older dl4j api. If there’s “new network k + 1” including activation functions usually building blocks should be available for that.

I can imagine it is possible, just wondering if anyone has done it, in order to avoid double work. I am not sure what you mean with “new network k +1”

KAN look very promising and offer a fresh look to ML

Where can I find related sample code?

@cqiaoYc deeplearning4j-examples/samediff-examples at master · deeplearning4j/deeplearning4j-examples · GitHub as well as in the unit tests under platform-tests in the source code. Feel free to search SameDiff in there too.