Skip to yearly menu bar Skip to main content

Workshop: 4th Workshop on Self-Supervised Learning: Theory and Practice

Self-Supervised Disentanglement by Leveraging Structure in Data Augmentations

Cian Eastwood · Julius von Kügelgen · Linus Ericsson · Diane Bouchacourt · Pascal Vincent · Bernhard Schölkopf · Mark Ibrahim


Self-supervised representation learning often uses data augmentations to induce some invariance to “style” attributes of the data. However, with downstream tasks generally unknown at training time, it is difficult to deduce a priori which attributes of the data are indeed “style” and can be safely discarded. To address this, we introduce a more principled approach that seeks to disentangle style features rather than discard them. The key idea is to add multiple style embedding spaces where: (i) each is invariant to all-but-one augmentation; and (ii) joint entropy is maximized. We empirically demonstrate the benefits of our approach on synthetic datasets and then present promising but limited results on ImageNet.

Chat is not available.