Leveraging tensor kernels to reduce objective function mismatch in deep clustering

Daniel J. Trosten, Sigurd Løkse, Robert Jenssen, Michael Kampffmeyer

Abstract: Highlights•Autoencoder-based deep clustering is prone to suffer from objective function mismatch.•We present the unsupervised companion objectives to reduce this mismatch.•The companion objectives use tensor kernels to measure representation similarities.•Experiments demonstrate the efficacy of the proposed methodology.