Timezone: »

 
Poster
Contrastive Learning Inverts the Data Generating Process
Roland S. Zimmermann · Yash Sharma · Steffen Schneider · Matthias Bethge · Wieland Brendel

Wed Jul 21 09:00 PM -- 11:00 PM (PDT) @ Virtual

Contrastive learning has recently seen tremendous success in self-supervised learning. So far, however, it is largely unclear why the learned representations generalize so effectively to a large variety of downstream tasks. We here prove that feedforward models trained with objectives belonging to the commonly used InfoNCE family learn to implicitly invert the underlying generative model of the observed data. While the proofs make certain statistical assumptions about the generative model, we observe empirically that our findings hold even if these assumptions are severely violated. Our theory highlights a fundamental connection between contrastive learning, generative modeling, and nonlinear independent component analysis, thereby furthering our understanding of the learned representations as well as providing a theoretical foundation to derive more effective contrastive losses.

Author Information

Roland S. Zimmermann (University of Tübingen, International Max Planck Research School for Intelligent Systems)
Yash Sharma (University of Tübingen)
Steffen Schneider (University of Tübingen)
Matthias Bethge (University of Tübingen)
Wieland Brendel (University of Tübingen)

Related Events (a corresponding poster, oral, or spotlight)

More from the Same Authors