1 paper
Jaeyoung Kim, Eunseok Kim, Dongsuk Jang
Hyperbolic vision-language models are designed to encode abstraction geometrically: general concepts near the origin, specific ones farther out, and entailment cones representing d…