Evaluating Federated Kolmogorov-Arnold Networks on Non-IID Data

TL;DR

A comparison between KANs and Multi-Layer Perceptrons with a similar number of parameters for 100 rounds of federated learning in the MNIST classification task using non-IID partitions with 100 clients shows that the best accuracies achieved by MLPs can be achieved by Spline-KANs in half of the time.

Abstract

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.

Figures (3)

• Figure 1: Per partition labels distribution.
• Figure 2: Mean test accuracy for 100 rounds, with error bands.
• Figure 3: Mean accuracy and loss metrics for 100 rounds, with error bands.