|
|
|
|
|
|
by Mahesh_3
775 days ago
|
|
|
Yes, i agree with that. Kolmogorov-Arnold theorems most work has stuck with the original depth-2 width-(2n + 1) representation, and did not have the chance to leverage more modern techniques (Like.. back propagation techinque) to train the networks. The biggest bottleneck of KANs lies in its slow training.from the research paper they had observed that KANs are usually 10x slower than MLPs, given the same number of parameters. So, KANs’ slow training is more of a engineering problem to be improved in the future rather than a fundamental limitation. It's more practical for ML Infrastructure industry leaders to take charge of solving this engineering problem. umamaheswar edara |
|
|