Large-margin representation learning for texture classification

This paper presents a novel approach combining convolutional layers (CLs) and large-margin metric learning for training supervised models on small datasets texture classification. The core of such an is loss function that computes the distances between instances interest support vectors. objective to update weights CLs iteratively learn representation with large margin classes. Each iteration results in discriminant model represented by vectors based representation. advantage proposed w.r.t. neural networks (CNNs) two-fold. First, it allows amount data due reduced number parameters compared equivalent CNN. Second, has low cost since backpropagation considers only experimental histopathologic image show achieves competitive accuracy lower computational faster convergence CNNs.