{
  "id": 6846,
  "url": "https://arxiv.org/abs/2603.22042v3",
  "title": "Uncertainty-guided Compositional Alignment with Part-to-Whole Semantic Representativeness in Hyperbolic Vision-Language Models",
  "summary": "While Vision-Language Models (VLMs) have achieved remarkable performance, their Euclidean embeddings remain limited in capturing hierarchical relationships such as part-to-whole or parent-child structures, and often face challenges in multi-object compositional scenarios. Hyperbolic VLMs mitigate this issue by better preserving hierarchical structures and modeling part-whole relations (i.e., whole scene and its part images) through entailment. However, existing approaches do not model that each ",
  "authors": "Hayeon Kim, Ji Ha Jang, Junghun James Kim, Se Young Chun",
  "category": "research",
  "topics": "safety-alignment,children-education",
  "orgs": null,
  "regions": null,
  "published_at": "2026-03-23T14:41:20.000Z",
  "fetched_at": "2026-07-14T16:32:45.896Z",
  "source_slug": "arxiv-ethics",
  "source_name": "arXiv",
  "source_homepage": "https://arxiv.org",
  "ethics_ai_record_url": "https://ethics.ai/record/6846",
  "original_url": "https://arxiv.org/abs/2603.22042v3",
  "evidence_status": "source-only",
  "attribution": "via ethics.ai"
}