BSRBF-KAN: A mixture of B-splines and Radial Primary Capabilities in Kolmogorov-Arnold Networks
Authors: Hoang-Thang Ta
Summary: On this paper, we introduce BSRBF-KAN, a Kolmogorov Arnold Community (KAN) that mixes Bsplines and radial foundation features (RBFs) to suit enter vectors in information coaching. We carry out experiments with BSRBF-KAN, MLP, and different well-liked KANs, together with EfficientKAN, FastKAN, FasterKAN, and GottliebKAN over the MNIST dataset. BSRBF-KAN exhibits stability in 5 coaching instances with a aggressive common accuracy of 97.55% and obtains convergence higher than different networks. We anticipate BSRBF-KAN can open many mixtures of mathematical features to design KANs. Our repo is publicly out there at: https://github.com/hoangthangta/BSRBF-KAN.
