arXiv Releases SW-KAN: Kolmogorov-Arnold Networks with Stieltjes-Wigert q-Orthogonal Polynomials
Summary:arXiv has introduced the Stieltjes-Wigert Kolmogorov-Arnold Network (SW-KAN), a novel deep learning architecture that addresses the domain mismatch issue in traditional polynomial-based KANs by employing Stieltjes-Wigert q-orthogonal polynomials defined on the semi-infinite domain (0, infinity). The architecture introduces a smooth exponential-of-tanh mapping to bridge the domain gap while preserving well-conditioned gradients and leverages a numerically stable three-term recurrence for efficien
A New Deep Learning Architecture: SW-KAN
Kolmogorov-Arnold Networks (KANs) represent a paradigm shift in deep learning by replacing fixed node activations with learnable univariate functions on edges, offering enhanced interpretability and parameter efficiency. However, existing polynomial-based KAN variants, while addressing the computational overhead of original B-spline implementations, introduce a fundamental challenge: the domain mismatch between unbounded real-valued inputs and the bounded or semi-infinite support of orthogonal polynomial bases.
To tackle this issue, arXiv has introduced the Stieltjes-Wigert Kolmogorov-Arnold Network (SW-KAN), featuring the following innovations:
- Stieltjes-Wigert q-orthogonal polynomials: These polynomials, defined on the semi-infinite domain (0, infinity), resolve the domain mismatch problem inherent in traditional polynomial bases.
- Smooth exponential-of-tanh mapping: This mapping stably bridges the domain gap while preserving well-conditioned gradients.
- Numerically stable three-term recurrence: The recurrence formula enables efficient evaluation of polynomial expansions in O(N) operations without the need for special-function calls.
Experimental Results and Performance
In comprehensive experiments spanning image classification and continuous function approximation, SW-KAN demonstrates superior performance:
- Accuracy-efficiency trade-off: SW-KAN achieves a balanced trade-off between accuracy and efficiency, particularly excelling in resource-constrained environments.
- Robustness: The log-normal weight structure and learnable q-parameter provide a distinct inductive bias, enabling robust performance even with limited feature dimensionality and training data.
Technical Highlights and Industry Impact
- Innovative solution to domain mismatch: SW-KAN offers a novel approach to handling unbounded real-valued inputs, expanding the applicability of KANs.
- Efficient polynomial evaluation: The numerical stability of the recurrence formula makes SW-KAN suitable for large-scale data processing.
- Applicability in resource-constrained settings: The architecture's efficiency makes it ideal for edge computing and embedded systems.
Recommendations for Developers
- Experiment with SW-KAN: For tasks involving unbounded real-valued inputs, developers should consider experimenting with SW-KAN to leverage its domain handling capabilities and robustness.
- Focus on resource efficiency: The efficiency of SW-KAN is particularly beneficial in resource-constrained environments, making it a strong candidate for such applications.
- Explore broader applications: Beyond image classification and function approximation, the potential of SW-KAN in other domains such as time series prediction and natural language processing is worth exploring.
— END —Source: ArXiv Machine Learning (cs.LG) (2026-10-02)
Tags: #Deep Learning #Kolmogorov-Arnold Networks #Stieltjes-Wigert q-Orthogonal Polynomial #Resource-Constrained Computing #Efficient Function Approximation
Editorial & Fact-Checking Note: This article is compiled from primary research, official release documentation, and source papers by the ZICQ Newsroom pipeline with automated entity verification and human editorial review. If you notice any technical inaccuracy, please submit a correction via our corrections policy or email our editorial desk directly.
Community Comments