Evaluating Federated Kolmogorov-Arnold Networks on Non-IID Data
This work provides an incremental evaluation for researchers in federated learning, comparing F-KANs to MLPs on non-IID data.
The paper tackled the problem of evaluating Federated Kolmogorov-Arnold Networks (F-KANs) on non-IID data in federated learning, showing that Spline-KANs achieved the same best accuracies as MLPs in half the number of rounds, with a moderate increase in computing time.
Federated Kolmogorov-Arnold Networks (F-KANs) have already been proposed, but their assessment is at an initial stage. We present a comparison between KANs (using B-splines and Radial Basis Functions as activation functions) and Multi- Layer Perceptrons (MLPs) with a similar number of parameters for 100 rounds of federated learning in the MNIST classification task using non-IID partitions with 100 clients. After 15 trials for each model, we show that the best accuracies achieved by MLPs can be achieved by Spline-KANs in half of the time (in rounds), with just a moderate increase in computing time.