Here are what I think are the main conclusions of the article: """ ... the most significant factor controlling performance is just parameter count. """ """ No

Here are what I think are the main conclusions of the article: """ ... the most significant factor controlling performance is just parameter count. """ """ No matter what I did, the most simple neural network was still outperforming the fanciest KAN-based model I tried. """ I suspected this was the case when I first heard about KANs. Its nice to see someone diving into a bit more, even if it is just anecdotal.