We are proud to announce that our paper "Details (Don't) Matter: Isolating Cluster Information in Deep Embedded Spaces" by Lukas Miklautz* (University of Vienna), Lena Greta-Marie Bauer* (University of Vienna), Dominik Mautz (Ludwig-Maximilians-Universität), Sebastian Tschiatschek (University of Vienna), Christian Böhm (Ludwig-Maximilians-Universität), and Claudia Plant (University of Vienna) has been accepted for presentation at IJCAI-21 (the 30th International Joint Conference on Artificial Intelligence). Out of the 4204 full-paper submissions, 587 papers are finally accepted, at a 13.9% acceptance rate.
We proposed our deep clustering framework, ACe/DeC, that is compatible with Autoencoder Centroid based Deep Clustering methods and automatically learns a latent representation consisting of two separate spaces. The clustering space captures all cluster-specific information and the shared space explains general variation in the data. This resolves the conflict of existing deep clustering methods between preserving details, as expressed by the reconstruction loss, and finding similar groups by ignoring details, as expressed by the clustering loss.
∗ First authors with equal contribution.