Skip to yearly menu bar Skip to main content


Poster

SODA: Bottleneck Diffusion Models for Representation Learning

Drew Hudson · Daniel Zoran · Mateusz Malinowski · Andrew Lampinen · Andrew Jaegle · James McClelland · Loic Matthey · Felix Hill · Alexander Lerchner

Arch 4A-E Poster #348
[ ] [ Paper PDF ]
Fri 21 Jun 10:30 a.m. PDT — noon PDT

Abstract:

We introduce SODA, a self-supervised diffusion model, designed for representation learning. The model incorporates an image encoder, which distills a source view into a compact representation, that, in turn, guides the generation of related novel views. We show that by imposing a tight bottleneck between the encoder and a denoising decoder, and leveraging novel view synthesis as a self-supervised objective, we can turn diffusion models into strong representation learners, capable of capturing visual semantics in an unsupervised manner. To the best of our knowledge, SODA is the first diffusion model to succeed at ImageNet linear-probe classification, and, at the same time, it accomplishes reconstruction, editing and synthesis tasks across a wide range of datasets. Further investigation reveals the disentangled nature of its emergent latent space, that serves as an effective interface to control and manipulate the produced images. All in all, we aim to shed light on the exciting and promising potential of diffusion models, not only for image generation, but also for learning rich and robust representations. See our website at soda-diffusion.github.io.

Chat is not available.