Skip to content
AI News HubLIVE
Original source2 min read

Provable Cross-Modal Transfer of Hierarchical Concept Structure in CLIP Embeddings

Summary

The paper formalizes Hierarchical Angular Separation (HAS) observed in CLIP embeddings and proves that, in an unconstrained features model with fixed image embeddings, text-side HAS violation is bounded by image-side violation, driven by hierarchical co-occurrence of image-text pairs.

SourcearXiv Computer VisionAuthor: Nghia Nguyen, Tianjiao Ding, Lachlan Ewen MacDonald, Ren\'e Vidal
Provable Cross-Modal Transfer of Hierarchical Concept Structure in CLIP Embeddings
Report an error

The correction channel is not available yet. You can copy the article reference below for later.

Correction instructions
Read article

[Submitted on 13 Nov 2026]

Title:Provable Cross-Modal Transfer of Hierarchical Concept Structure in CLIP Embeddings

View a PDF of the paper titled Provable Cross-Modal Transfer of Hierarchical Concept Structure in CLIP Embeddings, by Nghia Nguyen and 3 other authors

View PDF HTML (experimental)

Abstract:Contrastive vision-language models such as CLIP are trained to align images and text, not to represent taxonomies. Yet empirical studies show that CLIP embeddings exhibit hierarchical structure aligned with the hierarchy of semantic concepts. A theoretical characterization of why such hierarchy emerges from a non-hierarchical objective remains open. In this paper, we formalize this phenomenon as Hierarchical Angular Separation (HAS), a geometric property whereby each non-leaf concept embedding is more similar to its descendants than to off-branch concepts. Empirically, we show that HAS holds approximately in raw images and randomly initialized CLIP image embeddings, while in text embeddings it holds approximately only after contrastive learning. To theoretically explain this asymmetry, we analyze an unconstrained features model in which the image embeddings are fixed and only the text embeddings are trained. At the global minimizer, the text-side HAS violation percentage at each node is at most the image-side HAS violation and near-violation percentage. The key mechanism is the hierarchical co-occurrence structure of image-text pairs, which biases text embeddings toward root-centered image prototypes. For a linear contrastive loss, we derive a closed-form solution that makes this mechanism explicit. We extend the analysis to the InfoNCE loss and establish analogous guarantees under sufficient regularization. Synthetic experiments validate the theory and show that HAS is transferred even beyond the regimes covered by our analysis.

Comments: NeurIPS 2026

Subjects:

Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG)

Cite as: arXiv:2611.17671 [cs.CV]

(or arXiv:2611.17671v1 [cs.CV] for this version)

https://doi.org/10.48550/arXiv.2611.17671

arXiv-issued DOI via DataCite (pending registration)

Submission history

From: Nghia Nguyen [view email] [v1] Fri, 13 Nov 2026 02:34:52 UTC (755 KB)

Full-text links:

Access Paper:

View a PDF of the paper titled Provable Cross-Modal Transfer of Hierarchical Concept Structure in CLIP Embeddings, by Nghia Nguyen and 3 other authors

View PDF

HTML (experimental)

TeX Source

view license

Additional Features

Audio Summary

Current browse context:

cs.CV

new | recent | 2026-11

Change to browse by:

cs cs.LG

References & Citations

SciX

Google Scholar

Semantic Scholar

Loading...

Data provided by:

Bibliographic Tools

Bibliographic and Citation Tools

Bibliographic Explorer Toggle

Bibliographic Explorer (What is the Explorer?)

Connected Papers Toggle

Connected Papers (What is Connected Papers?)

Litmaps Toggle

Litmaps (What is Litmaps?)

scite.ai Toggle

scite Smart Citations (What are Smart Citations?)

Code, Data, Media

Code, Data and Media Associated with this Article

alphaXiv Toggle

alphaXiv (What is alphaXiv?)

Links to Code Toggle

CatalyzeX Code Finder for Papers (What is CatalyzeX?)

DagsHub Toggle

DagsHub (What is DagsHub?)

GotitPub Toggle

Gotit.pub (What is GotitPub?)

Huggingface Toggle

Hugging Face (What is Huggingface?)

ScienceCast Toggle

ScienceCast (What is ScienceCast?)

Demos

Demos

Replicate Toggle

Replicate (What is Replicate?)

Spaces Toggle

Hugging Face Spaces (What is Spaces?)

Spaces Toggle

TXYZ.AI (What is TXYZ.AI?)

Related Papers

Recommenders and Search Tools

Link to Influence Flower

Influence Flower (What are Influence Flowers?)

Core recommender toggle

CORE Recommender (What is CORE?)

Author

Venue

Institution

Topic

About arXivLabs

arXivLabs: experimental projects with community collaborators

arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.

Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.

Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.

Which authors of this paper are endorsers? | Disable MathJax (What is MathJax?)

Key points and analysis

Article intelligence

EngineersAdvanced

Key points

  • Introduces HAS geometric property: non-leaf concept embeddings are more similar to descendants than off-branch concepts.
  • Experiments show HAS holds approximately in raw images and randomly initialized CLIP image embeddings, but in text embeddings only after contrastive learning.
  • Theory shows text-side HAS violation percentage is at most image-side violation in a model with fixed image embeddings.
  • Key mechanism is hierarchical co-occurrence structure of image-text pairs, biasing text embeddings toward root-centered prototypes.

Highlights and analysis are generated automatically and may contain errors. Check the original source.