Randomized non-linear PCA networks

PCANet is an unsupervised Convolutional Neural Network (CNN), which uses Principal Component Analysis (PCA) to learn convolutional filters. One drawback of that linear PCA cannot capture nonlinear structures within data. To address this problem, a straightforward approach utilizing kernel methods by equipping the method in with function. However, practice leads network having cubic complexity respect number training image patches. In paper, we propose called Randomized Nonlinear (RNPCANet), explicit Although RNPCANet utilizes for processing data, using approximation techniques define feature space each stage, theoretically show model not much higher than PCANet. We also our links PCANets Kernel Networks (CKNs) as proposed maps patches similar CKNs. evaluate on recognition tasks including Coil-20, Coil-100, ETH-80, Caltech-101, MNIST, and C-Cube datasets. The experimental results has superiority over CKNs terms accuracy.