A Deep Unsupervised Representation Learning Architecture using Coefficient of Variation in Hidden Layers

The aim of this paper is to present a novel cov-RBM model on the basis of Restricted Boltzmann Machine (RBM) and the coefficient of variation (CoV), where the learning process is instructed by the CoV eigenvalues with respect to the hidden layer. Further, a deep unsupervised representation learning architecture (DURLA) is proposed to explore the capacity for representation learning with continuous excitation of CoV eigenvalues. We subject the proposed DURLA to extensive experimentation on nine real-valued datasets. The final experimental results show that our methods have a better effect than traditional and state-of-the-art ways.