Abstract:This paper introduces a novel application of Kolmogorov-Arnold Networks (KANs) to time series forecasting, leveraging their adaptive activation functions for enhanced predictive modeling. Inspired by the Kolmogorov-Arnold representation theorem, KANs replace traditional linear weights with spline-parametrized univariate functions, allowing them to learn activation patterns dynamically. We demonstrate that KANs outperforms conventional Multi-Layer Perceptrons (MLPs) in a real-world satellite traffic forecasting task, providing more accurate results with considerably fewer number of learnable parameters. We also provide an ablation study of KAN-specific parameters impact on performance. The proposed approach opens new avenues for adaptive forecasting models, emphasizing the potential of KANs as a powerful tool in predictive analytics.

What problem does this paper attempt to address?

### What problem does this paper attempt to solve? This paper primarily explores the application of Kolmogorov-Arnold Networks (KANs) in time series forecasting and evaluates their advantages over traditional Multi-Layer Perceptrons (MLPs). #### Specific Objectives: 1. **Introduction of KANs**: Propose a new neural network architecture—Kolmogorov-Arnold Networks (KANs), utilizing their adaptive activation functions to enhance predictive modeling. 2. **Performance Comparison**: Demonstrate the superiority of KANs in prediction accuracy through real-world tasks involving satellite traffic data, especially with fewer parameters. 3. **Parameter Analysis**: Conduct an ablation study on the impact of specific parameters of KANs, exploring the effects of the number of nodes and grid size on model performance. 4. **Theoretical Foundation**: Based on the Kolmogorov-Arnold representation theorem, KANs replace traditional linear weights with univariate functions parameterized by splines, thereby dynamically learning activation patterns. 5. **Practical Application**: Validate the effectiveness of KANs in time series forecasting using real-world satellite traffic datasets. ### Main Findings: 1. **Performance Improvement**: KANs outperform traditional MLPs in prediction accuracy, showing higher flexibility and accuracy, especially in handling sudden changes. 2. **Parameter Efficiency**: KANs have a lower number of parameters, achieving higher prediction accuracy with fewer trainable parameters. 3. **Configuration Optimization**: Increasing the number of nodes and grid size can significantly enhance the performance of KANs, but it requires a balance with computational resource demands. ### Conclusion: KANs demonstrate significant advantages in time series forecasting, particularly in applications like satellite traffic prediction, showcasing their potential as an alternative to traditional MLPs. However, further research is needed to optimize their performance in a broader range of applications.

Kolmogorov-Arnold Networks (KANs) for Time Series Analysis

Kolmogorov-Arnold Networks for Time Series: Bridging Predictive Power and Interpretability

Kolmogorov-Arnold Networks (KAN) for Time Series Classification and Robust Analysis

SigKAN: Signature-Weighted Kolmogorov-Arnold Networks for Time Series

KAN: Kolmogorov-Arnold Networks

Exploring Kolmogorov-Arnold Networks for Interpretable Time Series Classification

KAN-AD: Time Series Anomaly Detection with Kolmogorov-Arnold Networks

C-KAN: A New Approach for Integrating Convolutional Layers with Kolmogorov–Arnold Networks for Time-Series Forecasting

TKAN: Temporal Kolmogorov-Arnold Networks

A Revolutionary Neural Network Architecture with Interpretability and Flexibility Based on Kolmogorov–Arnold for Solar Radiation and Temperature Forecasting

A Survey on Kolmogorov-Arnold Network

KAN4TSF: Are KAN and KAN-based models Effective for Time Series Forecasting?

A Comprehensive Survey on Kolmogorov Arnold Networks (KAN)

KKANs: Kurkova-Kolmogorov-Arnold Networks and Their Learning Dynamics

A Benchmarking Study of Kolmogorov-Arnold Networks on Tabular Data

Convolutional Kolmogorov-Arnold Networks

On the expressiveness and spectral bias of KANs

SineKAN: Kolmogorov-Arnold Networks Using Sinusoidal Activation Functions

Smooth Kolmogorov Arnold networks enabling structural knowledge representation

Kolmogorov-Arnold Networks in Low-Data Regimes: A Comparative Study with Multilayer Perceptrons

A Novel Interpretable Short-Term Load Forecasting Method Based on Kolmogorov-Arnold Networks