DropKAN: Dropout Kolmogorov-Arnold Networks

We propose DropKAN (Dropout Kolmogorov-Arnold Networks) a regularization method that prevents co-adaptation of activation function weights in Kolmogorov-Arnold Networks (KANs). DropKAN functions by embedding the drop mask directly within the KAN layer, randomly masking the outputs of some activations within the KANs’ computation graph. We show that this procedure has a regularizing effect and consistently lead to better generalization of KANs. We analyze the adaptation of the standard Dropout with KANs and demonstrate that Dropout applied to KANs’ neurons can lead to unpredictable behavior in the feedforward pass.We carry an extensive empirical study with real-world Machine Learning datasets and computer vision benchmarks to validate our findings. Our results suggest that DropKAN is consistently a better alternative to using standard Dropout with KANs, and improves the generalization performance of KANs even under high drop rates or limited training g data. Our implementation of DropKAN is available a .