An Incremental Deep Learning Network for On-line Unsupervised Feature Extraction

Yu Liang, Yi Yang, Furao Shen, Jinxi Zhao, Tao Zhu

Abstract: In this paper, we propose an incremental deep learning network for on-line unsupervised feature extraction. This deep learning network is based on 3 data processing components: (1) cascaded incremental orthogonal component analysis network (IOCANet); (2) binary hashing; and (3) blockwise histograms. In this architecture, IOCANet can process online data and get filters to do convolutions. Binary hashing is used to enhance the nonlinearity of IOCANet and reduce the quantity of the data. Eventually, the data is encoded by blockwise histograms. Experiments demonstrate that the proposed architecture has potential results for on-line unsupervised feature extraction.