Hyperbolic Concept Bottleneck Models

Daniel Uyterlinde, Swasti S. Mishra, Pascal Mettes · 2026

Read Hyperbolic Concept Bottleneck Models by Daniel Uyterlinde, Swasti S. Mishra, Pascal Mettes (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2605-06440.txt.

Description

Daniel Uyterlinde, Swasti Shreya Mishra, and Pascal Mettes propose putting concept bottleneck models into curved hyperbolic space. Such models force a classifier to decide through human-understandable concepts, which aids interpretation. Standard versions place concepts in flat space and treat them as independent, yet real concepts form hierarchies.

Their HypCBM treats a concept's activation as containment inside an entailment cone, so broader ideas geometrically contain narrower ones. The inclusion margin gives hierarchy-aware activations at test time with no extra learned parts. A scaling rule also spreads user corrections through the concept tree, and tests rival Euclidean models trained on far more data.

Catalog record

CREATORDaniel Uyterlinde, Swasti S. Mishra, Pascal Mettes
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE65 KB
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2605.06440

Browse the full library · About the Tower of Babel