Learnable Subspace Orthogonal Projection for Semi-supervised Image Classification.

In this paper, we propose a learnable subspace orthogonal projection (LSOP) network for semi-supervised image classification. Although projection theory is widely used in various machine learning methods, solving projection matrix is a highly complex process. We employ an auto-encoder to construct a scalable and learnable subspace orthogonal projection network, thus enjoying lower computational consumption of subspace acquisition and smooth cooperation with deep neural networks. With these techniques, a promising end-to-end classification network is formulated. Extensive experimental results on real-world datasets demonstrate that the proposed classification algorithm achieves comparable performance with fewer training data than other projection methods.