Learning State-Space Models with Kolmogorov-Arnold Networks: an Autonomous Underwater Engineering perspective

Learning State-Space Models with Kolmogorov-Arnold Networks: An Autonomous Underwater Engineering Perspective This Zenodo repository contains the source code, experimental results, and test logs associated with the paper: “Learning State-Space Models with Kolmogorov-Arnold Networks: An Autonomous Underwater Engineering Perspective.” The repository includes the material produced for the initial version of the manuscript, as well as the additional experiments conducted in response to the reviewers’ comments. Repository Contents The repository contains the following three ZIP archives. SOURCE_CODE_&_TESTS_V1.zip This archive contains: The source code used for the first revision of the paper. The complete test logs and experimental outputs associated with the experiments presented in the first revision. The scripts and configurations used to train and evaluate the considered models. This archive is intended to preserve the original experimental setup and results associated with the initial submission. SOURCE_CODE_V2.zip This archive contains: The source code used for the second revision of the paper. The scripts and configurations used to train and evaluate the considered models. AUV_TESTS_v2.zip This archive contains the additional Autonomous Underwater Vehicle, or AUV, experiments performed after the first review. In particular, it includes: Additional tests requested by the reviewers. Integrated experiments using Chebyshev KAN architectures for the AUV case. Integrated experiments using Jacobi KAN architectures for the AUV case. The corresponding logs, outputs, and evaluation results. These experiments extend the original analysis by considering additional Kolmogorov-Arnold Network variants in the AUV system-identification setting. AUV_SYMBOLIC_v2.zip This archive contains the additional Autonomous Underwater Vehicle, or AUV, experiments performed after the first review related only to the symbolification phase. In particular, it includes: Additional tests requested by the reviewers. The corresponding logs, outputs, and evaluation results. These experiments extend the original analysis by extending model symbolification results. CLASSICAL_TESTS_V2.zip This archive contains the classical benchmark experiments that were repeated after the first review. The experiments were re-executed using networks with comparable numbers of parameters. This parameter-matching procedure was introduced to enable a fairer comparison between the different architectures and to provide a clearer understanding of their performance across the considered test cases. The archive includes: Re-executed classical benchmark experiments. Parameter-matched model configurations. Training and evaluation logs. Experimental outputs and performance results. Purpose of the Additional Experiments The experiments included in AUV_TESTS_v2.zip and CLASSICAL_TESTS_V2.zip were conducted in response to the comments received during the first review round. Their main purposes are to: Extend the AUV analysis to Chebyshev and Jacobi KAN architectures. Improve the fairness of the comparisons between different models. Ensure that the compared networks have similar numbers of parameters. Better assess model performance under different system-identification scenarios. Provide additional evidence supporting the analysis and conclusions of the revised paper. Reproducibility The archives provide the source code, test configurations, logs, and experimental results required to inspect and reproduce the analyses discussed in the paper. Users should refer to the documentation and scripts contained in each archive for experiment-specific execution instructions, dependencies, dataset paths, and model configurations.