Abstract:Time series classification is a relevant step supporting decision-making processes in various domains, and deep neural models have shown promising performance. Despite significant advancements in deep learning, the theoretical understanding of how and why complex architectures function remains limited, prompting the need for more interpretable models. Recently, the Kolmogorov-Arnold Networks (KANs) have been proposed as a more interpretable alternative. While KAN-related research is significantly rising, to date, the study of KAN architectures for time series classification has been limited. In this paper, we aim to conduct a comprehensive and robust exploration of the KAN architecture for time series classification on the UCR benchmark. More specifically, we look at a) how reference architectures for forecasting transfer to classification, at the b) hyperparameter and implementation influence on the classification performance in view of finding the one that performs best on the selected benchmark, the c) complexity trade-offs and d) interpretability advantages. Our results show that (1) Efficient KAN outperforms MLP in performance and computational efficiency, showcasing its suitability for tasks classification tasks. (2) Efficient KAN is more stable than KAN across grid sizes, depths, and layer configurations, particularly with lower learning rates. (3) KAN maintains competitive accuracy compared to state-of-the-art models like HIVE-COTE2, with smaller architectures and faster training times, supporting its balance of performance and transparency. (4) The interpretability of the KAN model aligns with findings from SHAP analysis, reinforcing its capacity for transparent decision-making.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is: **Explore the application and interpretability of Kolmogorov - Arnold Networks (KANs) in time - series classification**. Specifically, the author hopes to improve the performance and interpretability of time - series classification tasks by using the KAN architecture. ### Main research objectives: 1. **Transfer of existing prediction architectures**: Evaluate whether the existing KAN architecture (originally used for time - series prediction) can be effectively applied to time - series classification tasks. 2. **Influence of hyper - parameters and implementation**: Analyze the influence of different hyper - parameters (such as grid size, network depth, node configuration, etc.) on classification performance to find the configuration that performs best on the selected benchmark. 3. **Trade - off between complexity and interpretability**: Explore the balance between complexity and interpretability in the KAN model, especially in terms of computational efficiency and transparency. ### Specific problems: - **How to transfer existing prediction architectures to classification tasks**: Research whether the design of the KAN architecture in prediction tasks can be directly or adjusted for use in classification tasks. - **Influence of hyper - parameters and implementation on classification performance**: By adjusting hyper - parameters such as grid size, network depth, and node configuration, evaluate how these factors affect classification performance. - **Complexity trade - off**: Compare the differences in computational complexity and performance among different models (such as the original KAN, Efficient KAN, and Multi - Layer Perceptron MLP). - **Interpretability advantage**: Verify the interpretability of the KAN model, especially the consistency between its feature importance and activation function and the SHAP analysis results. ### Main contributions of the paper: 1. **Suitability study**: Research whether the existing KAN architecture is suitable for time - series classification tasks on the UCR dataset. 2. **Hyper - parameter analysis**: By analyzing the influence of hyper - parameters on two KAN implementations, find the best KAN configuration. 3. **Performance and complexity comparison**: Compare the performance and computational complexity of the original KAN, Efficient KAN, and MLP in time - series classification tasks, and discover the advantages of Efficient KAN in terms of stability and computational efficiency. 4. **Interpretability confirmation**: Through in - depth analysis of the learned feature importance and activation function, verify the interpretability of the KAN model and compare it with the SHAP analysis results. In conclusion, this paper aims to explore the potential of the KAN architecture in time - series classification, especially its interpretability and performance, through comprehensive and robust research.

Exploring Kolmogorov-Arnold Networks for Interpretable Time Series Classification

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

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

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

Kolmogorov-Arnold Networks (KANs) for Time Series Analysis

A Survey on Kolmogorov-Arnold Network

Exploring the Limitations of Kolmogorov-Arnold Networks in Classification: Insights to Software Training and Hardware Implementation

Can KAN Work? Exploring the Potential of Kolmogorov-Arnold Networks in Computer Vision

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

On the expressiveness and spectral bias of KANs

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

KAN: Kolmogorov-Arnold Networks

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

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

HyperKAN: Kolmogorov–Arnold Networks Make Hyperspectral Image Classifiers Smarter

KAN we improve on HEP classification tasks? Kolmogorov-Arnold Networks applied to an LHC physics example

A Comprehensive Survey on Kolmogorov Arnold Networks (KAN)

Kolmogorov-Arnold Convolutions: Design Principles and Empirical Studies

HyperKAN: Kolmogorov-Arnold Networks make Hyperspectral Image Classificators Smarter

KANs for Computer Vision: An Experimental Study

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