Adaptive RBF-KAN improves efficiency with data-driven kernel initialization

Researchers have developed an enhanced version of Kolmogorov-Arnold Networks (KANs) called Adaptive RBF-KAN, which utilizes radial basis functions (RBFs) instead of traditional B-spline bases for improved computational efficiency. This new model introduces a broader family of RBF kernels and employs leave-one-out cross-validation (LOOCV) for data-driven initialization of kernel shape parameters. Evaluations on benchmark functions demonstrate that adaptive kernel selection and shape parameters significantly improve RBF-KAN performance, with different kernels showing advantages for various data characteristics like smoothness or discontinuities. AI

Summary written by gemini-2.5-flash-lite from 1 source. How we write summaries →

IMPACT Introduces a more efficient and adaptable neural network architecture, potentially improving performance on complex function approximation tasks.

RANK_REASON The cluster contains a new academic paper detailing a novel method for improving existing neural network architectures. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv stat.ML →

• Adaptive RBF-KAN
• arXiv
• B-spline
• FastKAN
• Gaussian radial basis functions
• Kolmogorov-Arnold Networks
• leave-one-out cross-validation
• Matérn kernels
• Wendland kernels

• paper
• model release