Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Fed-KAN: Federated Learning with Kolmogorov-Arnold Networks for Traffic Prediction

Engin Zeydan, Cristian J. Vaca-Rubio, Luis Blanco, Roberto Pereira, Marius Caus, Kapal Dev

TL;DR

The paper addresses traffic prediction in non-terrestrial networks (NTNs) under dynamic LEO conditions by proposing Fed-KAN, a federated learning framework that embeds Kolmogorov-Arnold Networks (KANs) as local models. Fed-KAN leverages spline-based, learnable activations within a FedAvg-style FL setup to improve function approximation and generalization while reducing ground-station communication. Empirical results on real satellite-beam data show Fed-KAN achieving a 77.39% reduction in average test loss compared with a Fed-MLP baseline, demonstrating faster convergence and better adaptation to NTN dynamics. The work also discusses deployment considerations within O-RAN architectures, including on-board vs ground-based processing and functional splits, highlighting practical pathways for privacy-preserving, low-latency NTN resource optimization.

Abstract

Non-Terrestrial Networks (NTNs) are becoming a critical component of modern communication infrastructures, especially with the advent of Low Earth Orbit (LEO) satellite systems. Traditional centralized learning approaches face major challenges in such networks due to high latency, intermittent connectivity and limited bandwidth. Federated Learning (FL) is a promising alternative as it enables decentralized training while maintaining data privacy. However, existing FL models, such as Federated Learning with Multi-Layer Perceptrons (Fed-MLP), can struggle with high computational complexity and poor adaptability to dynamic NTN environments. This paper provides a detailed analysis for Federated Learning with Kolmogorov-Arnold Networks (Fed-KAN), its implementation and performance improvements over traditional FL models in NTN environments for traffic forecasting. The proposed Fed-KAN is a novel approach that utilises the functional approximation capabilities of KANs in a FL framework. We evaluate Fed-KAN compared to Fed-MLP on a traffic dataset of real satellite operator and show a significant reduction in training and test loss. Our results show that Fed-KAN can achieve a 77.39% reduction in average test loss compared to Fed-MLP, highlighting its improved performance and better generalization ability. At the end of the paper, we also discuss some potential applications of Fed-KAN within O-RAN and Fed-KAN usage for split functionalities in NTN architecture.

Fed-KAN: Federated Learning with Kolmogorov-Arnold Networks for Traffic Prediction

TL;DR

The paper addresses traffic prediction in non-terrestrial networks (NTNs) under dynamic LEO conditions by proposing Fed-KAN, a federated learning framework that embeds Kolmogorov-Arnold Networks (KANs) as local models. Fed-KAN leverages spline-based, learnable activations within a FedAvg-style FL setup to improve function approximation and generalization while reducing ground-station communication. Empirical results on real satellite-beam data show Fed-KAN achieving a 77.39% reduction in average test loss compared with a Fed-MLP baseline, demonstrating faster convergence and better adaptation to NTN dynamics. The work also discusses deployment considerations within O-RAN architectures, including on-board vs ground-based processing and functional splits, highlighting practical pathways for privacy-preserving, low-latency NTN resource optimization.

Abstract

Non-Terrestrial Networks (NTNs) are becoming a critical component of modern communication infrastructures, especially with the advent of Low Earth Orbit (LEO) satellite systems. Traditional centralized learning approaches face major challenges in such networks due to high latency, intermittent connectivity and limited bandwidth. Federated Learning (FL) is a promising alternative as it enables decentralized training while maintaining data privacy. However, existing FL models, such as Federated Learning with Multi-Layer Perceptrons (Fed-MLP), can struggle with high computational complexity and poor adaptability to dynamic NTN environments. This paper provides a detailed analysis for Federated Learning with Kolmogorov-Arnold Networks (Fed-KAN), its implementation and performance improvements over traditional FL models in NTN environments for traffic forecasting. The proposed Fed-KAN is a novel approach that utilises the functional approximation capabilities of KANs in a FL framework. We evaluate Fed-KAN compared to Fed-MLP on a traffic dataset of real satellite operator and show a significant reduction in training and test loss. Our results show that Fed-KAN can achieve a 77.39% reduction in average test loss compared to Fed-MLP, highlighting its improved performance and better generalization ability. At the end of the paper, we also discuss some potential applications of Fed-KAN within O-RAN and Fed-KAN usage for split functionalities in NTN architecture.

Paper Structure

This paper contains 19 sections, 4 figures, 2 tables.

Table of Contents

  1. Introduction
  2. FL and KANs for NTN
  3. Motivation of the paper
  4. General Architecture
  5. Federated KAN
  6. Introduction to Fed-KAN
  7. Fed-KAN layers
  8. Evaluation Results
  9. Dataset
  10. Model Architecture and Parameters
  11. Implementation details
  12. Key Observations
  13. Fed-KAN Applications in NTNs
  14. Decision Engine and Actuator
  15. Discussions on Split Functions in O-RAN
  16. ...and 4 more sections

Figures (4)

  • Figure 1: General architecture of NTN and the existence of diverse application traffic on each separate beam of satellites.
  • Figure 2: Federated KAN Methodology applied within NTN scenario.
  • Figure 3: Fed-KAN classifier layers with input (uplink and downlink traffic of last $W$ hours), hidden and output layers (next hour percentage prediction of traffic categories).
  • Figure 4: Average Training MSE Loss over Fed-KAN rounds.