Abstract:In this study, we introduces a parameter-efficient model that outperforms traditional models in time series forecasting, by integrating High-order Polynomial Projection (HiPPO) theory into the Kolmogorov-Arnold network (KAN) framework. This HiPPO-KAN model achieves superior performance on long sequence data without increasing parameter count. Experimental results demonstrate that HiPPO-KAN maintains a constant parameter count while varying window sizes and prediction horizons, in contrast to KAN, whose parameter count increases linearly with window size. Surprisingly, although the HiPPO-KAN model keeps a constant parameter count as increasing window size, it significantly outperforms KAN model at larger window sizes. These results indicate that HiPPO-KAN offers significant parameter efficiency and scalability advantages for time series forecasting. Additionally, we address the lagging problem commonly encountered in time series forecasting models, where predictions fail to promptly capture sudden changes in the data. We achieve this by modifying the loss function to compute the MSE directly on the coefficient vectors in the HiPPO domain. This adjustment effectively resolves the lagging problem, resulting in predictions that closely follow the actual time series data. By incorporating HiPPO theory into KAN, this study showcases an efficient approach for handling long sequences with improved predictive accuracy, offering practical contributions for applications in large-scale time series data.

What problem does this paper attempt to address?

### Problems the Paper Attempts to Solve This paper aims to address several key issues in time series forecasting: 1. **Parameter Efficiency and Scalability**: - Traditional models experience a linear increase in the number of parameters as the window size increases when handling long sequence data, leading to a sharp rise in model complexity and computational cost. - This paper proposes a parameter-efficient model (HiPPO-KAN) that integrates the High-order Polynomial Projection (HiPPO) theory into the Kolmogorov-Arnold Network (KAN) framework, enabling the handling of long sequence data while keeping the number of parameters constant. 2. **Long-term Dependencies**: - Capturing long-term dependencies is crucial in time series analysis, especially in fields like finance, weather forecasting, and energy consumption, where data often contains patterns spanning long periods. - Traditional models like RNN and LSTM have limitations in handling long-term dependencies. HiPPO-KAN, through the effective capture of HiPPO theory, can better handle long-term dependencies and improve prediction accuracy. 3. **Lag Issue**: - Time series forecasting models often face the lag issue, where the model's predictions fail to promptly capture sudden changes in the data. - This paper addresses the lag issue by modifying the loss function to directly compute the mean squared error (MSE) of the coefficient vector in the HiPPO domain, enabling the model to respond more quickly to changes in the input data. ### Main Contributions 1. **Parameter Efficiency and Scalability**: - HiPPO-KAN demonstrates excellent parameter efficiency in univariate time series forecasting tasks. The dimension of the coefficient vector does not depend on the length of the input sequence, allowing the model to maintain a constant number of parameters when handling long sequence data, thus achieving scalability. 2. **Performance Improvement in Long-range Forecasting**: - Experimental results show that HiPPO-KAN significantly outperforms traditional KAN and other models specialized in handling sequence data (such as RNN and LSTM) in long-range forecasting tasks. By effectively capturing long-term dependencies through the HiPPO framework, it improves prediction accuracy. 3. **Innovative Combination of HiPPO Theory and KAN**: - Using HiPPO coefficients provides a concise and interpretable state representation of the time series system. When combined with the transparent architecture of KAN, this representation helps better understand and interpret the internal workings of the model. ### Conclusion By combining HiPPO theory with the KAN framework, the HiPPO-KAN model demonstrates significant advantages in time series forecasting, particularly in handling long sequence data and capturing long-term dependencies. Additionally, by modifying the loss function, the model effectively addresses the lag issue, improving the timeliness and accuracy of predictions. These contributions provide practical methods and tools for the application of large-scale time series data.

HiPPO-KAN: Efficient KAN Model for Time Series Analysis

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

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

Kolmogorov-Arnold Networks (KANs) for Time Series Analysis

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

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

Exploring the power of KANs: Overcoming MLP limitations in complex data analysis

Hierarchical Proxy Modeling for Improved HPO in Time Series Forecasting

KANOP: A Data-Efficient Option Pricing Model using Kolmogorov-Arnold Networks

Exploring Kolmogorov-Arnold Networks for Interpretable Time Series Classification

HiFNet: rethinking time series forecasting models from a perspective of error power spectrum

KACDP: A Highly Interpretable Credit Default Prediction Model

Koopa: Learning Non-stationary Time Series Dynamics with Koopman Predictors

HiPPO: Recurrent Memory with Optimal Polynomial Projections

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

N-HiTS: Neural Hierarchical Interpolation for Time Series Forecasting

Predictive Modeling of Flexible EHD Pumps using Kolmogorov-Arnold Networks

Hierarchical Classification Auxiliary Network for Time Series Forecasting

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

How to Learn More? Exploring Kolmogorov-Arnold Networks for Hyperspectral Image Classification

TKAN: Temporal Kolmogorov-Arnold Networks