LSD-C: Linearly Separable Deep Clusters

Sylvestre-Alvise Rebuffi, Sebastien Ehrhardt, Kai Han, Andrea Vedaldi, Andrew Zisserman

Keywords: deep clustering, self-supervised learning, deep learning

TL;DR: We present a method to cluster unlabelled data points by subtly leveraging the pairwise connections in the feature space derived from self-supervision.

Abstract: We present LSD-C, a novel method to identify clusters in an unlabeled dataset. Our algorithm first establishes pairwise connections in the feature space between the samples of the minibatch based on a similarity metric. Then it regroups in clusters the connected samples and enforces a linear separation between clusters. This is achieved by using the pairwise connections as targets together with a binary cross-entropy loss on the predictions that the associated pairs of samples belong to the same cluster. This way, the feature representation of the network will evolve such that similar samples in this feature space will belong to the same linearly separated cluster. Our method draws inspiration from recent semi-supervised learning practice and proposes to combine our clustering algorithm with self-supervised pretraining and strong data augmentation. We show that our approach significantly outperforms competitors on popular public image benchmarks including CIFAR 10/100, STL 10 and MNIST, as well as the document classification dataset Reuters 10K. Our code is available at \url{https://www.dropbox.com/s/xnczi9ogw0ar87d/lsd-clusters.zip?dl=0}.

Community Implementations: [ 1 code implementation](https://www.catalyzex.com/paper/lsd-c-linearly-separable-deep-clusters/code)