Newest papers

Compositionality with Primitives: A Unified View from Euclidean to Hyperbolic Vision-Language Models

Puru Vaish, Meelad Dashti, Christoph Brüne, Pascal Mettes, Tatiana Tommasi, Nicola Strisciuglio · 2026 · arXiv (Cornell University)

DOI: 10.48550/arxiv.2610.29811 · cited by 0 works (OpenAlex) · open-access PDF

Portuguese

Composição zero-shot learning (CZSL) visa reconhecer novas combinações de objetos e atributos observados anteriormente. Trabalho anterior mostrou que modelos de linguagem de visão em larga escala codificam estruturas semânticas decompostáveis que podem ser recompostas para permitir a generalização. O framework Geodesically Decomposable Embeddings (GDE) formaliza isso identificando direções no espaço tangencial no frechet médio da distribuição embedded, que codificam conceitos visuais primitivos. Composições de atributos-objetos invisíveis podem ser sintetizadas através da adição de vetores desses primitivos. Revisamos a suposição de que a generalização de composição requer a re-composição explícita do grupo de atributos-objetos. Por meio de novas funções de pontuação

Machine translation (Meta M2M100); the English original follows.

Abstract

Compositional zero-shot learning (CZSL) aims to recognize novel combinations of previously observed objects and attributes. Prior work showed that large-scale vision-language models encode decomposable semantic structures that can be recomposed to enable generalization. The Geodesically Decomposable Embeddings (GDE) framework formalizes this by identifying directions in the tangent space at the Frechet mean of the embedding distribution, which encode primitive visual concepts. Unseen attribute-object compositions can be synthesized through vector addition of these primitives. We revisit the assumption that compositional generalization requires explicit attribute-object pair re-composition. By means of new compositional scoring functions and extensive evaluations, we show that directly using the decomposed primitives suffices for recognizing unseen compositions. In that, we gain insights on how the geometry of the embedding manifold (Euclidean, spherical, or hyperbolic) and dataset structure affect compositional reasoning. Across models (CLIP, DINOv2, MERU, HyCoCLIP), we show that compositional structure consistently emerges as primitive representations, enabling inference on CZSL benchmarks (MIT-States, VAW-CZSL, UT-Zappos50K, Clothing16K) and group robustness on Waterbirds and CelebA. Code: https://github.com/nis-research/Compositionality-with-Primitives.

Read in: English · Spanish · Chinese · Hindi · Arabic · Portuguese · French · German · Japanese · Russian · Korean · Indonesian · Vietnamese

Topics: Embedding, Principle of compositionality, ENCODE, Robustness (evolution), Manifold (fluid mechanics), Mathematics

Sources: OpenAlex (CC0), arXiv, Crossref. Agent Exchange