Population Risk Bounds for Kolmogorov-Arnold Networks Trained by DP-SGD with Correlated Noise

First population risk bounds established for Kolmogorov-Arnold Networks trained with mini-batch SGD and DP-SGD using correlated noise perturbations, extending KAN theory beyond full-batch gradient descent.