{
  "id": 12352,
  "url": "https://arxiv.org/abs/2607.18625v1",
  "title": "Norm or Direction? Decoding Vision Mambas for High-Resolution Vision",
  "summary": "Vision Mamba models replace quadratic self-attention with linear complexity selective state space models (SSMs), emerging as efficient visual backbones. However, MambaOut demonstrates that a Gated CNN block can match or exceed VMamba on image classification, questioning the necessity of SSMs for vision. This raises a fundamental question: do VMamba and MambaOut encode visual information differently at the representation level? To investigate, we apply cross model centered kernel alignment (CKA)",
  "authors": "Jin Yu, Juyoun Park",
  "category": "research",
  "topics": "safety-alignment,finance-investment",
  "orgs": null,
  "regions": null,
  "published_at": "2026-07-21T01:49:20.000Z",
  "fetched_at": "2026-07-22T05:10:49.469Z",
  "source_slug": "arxiv-ethics",
  "source_name": "arXiv",
  "source_homepage": "https://arxiv.org",
  "ethics_ai_record_url": "https://ethics.ai/record/12352",
  "original_url": "https://arxiv.org/abs/2607.18625v1",
  "evidence_status": "source-only",
  "attribution": "via ethics.ai"
}