TY - JOUR AU - Sucholutsky, Ilia AU - Schonlau, Matthias PY - 2018/12/24 Y2 - 2024/03/29 TI - ConvART: Improving Adaptive Resonance Theory for Unsupervised Image Clustering JF - Journal of Computational Vision and Imaging Systems JA - J. Comp. Vis. Imag. Sys. VL - 4 IS - 1 SE - Articles DO - 10.15353/jcvis.v4i1.326 UR - https://openjournals.uwaterloo.ca/index.php/vsl/article/view/326 SP - 3 AB - <p>While supervised learning techniques have become increasingly<br>adept at separating images into different classes, these techniques<br>require large amounts of labelled data which may not always be<br>available. We propose a novel neuro-dynamic method for unsuper-<br>vised image clustering by combining 2 biologically-motivated mod-<br>els: Adaptive Resonance Theory (ART) and Convolutional Neu-<br>ral Networks (CNN). ART networks are unsupervised clustering al-<br>gorithms that have high stability in preserving learned information<br>while quickly learning new information. Meanwhile, a major prop-<br>erty of CNNs is their translation and distortion invariance, which<br>has led to their success in the domain of vision problems. By<br>embedding convolutional layers into an ART network, the useful<br>properties of both networks can be leveraged to identify different<br>clusters within unlabelled image datasets and classify images into<br>these clusters. In exploratory experiments, we demonstrate that<br>this method greatly increases the performance of unsupervised<br>ART networks on a benchmark image dataset.</p> ER -