Matin Mahmood
Matin Mahmood is the founder of hyper³labs in Berlin, where he builds hierarchy-aware image and text embedding models for product and catalog search. His work focuses on hyperbolic representation learning, visual retrieval, and search systems for structured commercial data. He holds a joint degree in AI and Math from the University of Edinburgh.
Sessions
Hyperbolic embedding models for visual search
We trained an open image-text embedding model in hyperbolic space, so visual search can use hierarchy as well as similarity. The talk shows how radius and aperture create hierarchy inside the embedding model, how that helps retrieve specific items from compositional queries, and where the approach fails.