Autoencoders Learning Sparse Representation

Abstract

Many regularized autoencoders learn a sparse rep-resentation of data. This type of representation enhances robust-ness against noise and computational efficiencies. Our objective in this paper is to provide the conditions under which sparsity is encouraged by AE under a little less restrictive view of data. We have shown a relaxed observed representation of input data and given the conditions on AE to promote sparsity.