Fig. 2: Comparison of ConvKAN splines and CNN convolutional filters in 2D and 3D dimensions. | Communications Medicine

Fig. 2: Comparison of ConvKAN splines and CNN convolutional filters in 2D and 3D dimensions.

From: Volumetric spline-based Kolmogorov-Arnold architectures surpass CNNs, vision transformers, and graph networks for Parkinson’s disease detection

Fig. 2: Comparison of ConvKAN splines and CNN convolutional filters in 2D and 3D dimensions.

a 2D representation: The gridded square represents a single MRI slice. The colored 3 × 3 region within the grid illustrates an example CNN filter (red), while the green curve demonstrates a B-spline used in ConvKAN, with green dots indicating knots. b 3D representation: The cube represents a volumetric MRI scan. The red 3 × 3 × 3 region within the cube shows a CNN filter, while the curved surface represents a 3D B-spline used in ConvKAN, with green dots marking the knots.

Back to article page