Come to our poster @icmlconf.bsky.social (Thu 17, morning) to discover more about it Or check our paper: sthalles.github.io/sop/

Self-Organizing Visual Prototypes (SOP) is a new, non-parametric approach to SSL that addresses these issues by representing each region in feature space using multiple real data embeddings (support embeddings), rather than a single learnable prototype.

SSL, in particular, CL, relies more and more on prototypes (learnable representations that hold some group meaning) But! overclustering, enforcing equipartitions, and underrepresentation are still problems that plague the prototypes We solve them with our method: SOP

Journal discussion Tuesday!

Hello world!

Did you miss out on the latest Visual Intelligence Online Seminar on equivariant self-supervision by Aiden Durrant? 👨‍🏫 🤖 🧠 📺 You can watch Durrant's talk on the Visual Intelligence YouTube channel: youtu.be/bsUGgwDYdXQ?...