Paper Image

Kolmogorov-Arnold Networks for Satellite Traffic Forecasting

Published on:

14 May 2024

Primary Category:

Signal Processing

Paper Authors:

Cristian J. Vaca-Rubio,

Luis Blanco,

Roberto Pereira,

Màrius Caus

Bullets

Key Details

KANs use spline activation functions instead of linear weights for enhanced modeling

Tested on satellite traffic data, KANs outperformed MLPs in accuracy

KANs achieved better forecasts with 4x fewer trainable parameters than MLPs

Increasing KAN node counts and grid sizes improves performance at a computational cost

Results highlight KANs as an efficient alternative to MLPs for forecasting

AI generated summary

Kolmogorov-Arnold Networks for Satellite Traffic Forecasting

This paper proposes using Kolmogorov-Arnold Networks (KANs), a novel neural network architecture, for satellite traffic forecasting. KANs leverage spline-based activation functions that can learn complex patterns, replacing traditional linear weights. When tested on real-world satellite data, KANs significantly outperformed conventional Multi-Layer Perceptrons (MLPs), providing more accurate forecasts using far fewer trainable parameters. An ablation study also explores how KAN-specific parameters impact performance. Overall, this demonstrates the potential of KANs as an efficient and powerful tool for traffic forecasting tasks.

Answers from this paper

Comments

No comments yet, be the first to start the conversation...

Sign up to comment on this paper

Sign Up