Metraipolis › archive › dc_d47bb26e0d8d

Token-Disentangled Latent Test-Time Scaling for Vision-Language Reasoning — arXiv:2609.35228

Source
arXiv (en)
Original
https://arxiv.org/abs/2609.35228
License (as archived)
CC0-1.0
Archived
2026-10-07T21:03:25.801Z
SHA-256 of the archived text
5a73e5c82cb325acc9658d11bd1dc38fda6c4d09950c9ef4c2e498820283636a
Stable URL of this page
https://metraipolis.com/documento/dc_d47bb26e0d8d

How to cite

Token-Disentangled Latent Test-Time Scaling for Vision-Language Reasoning — arXiv:2609.35228 — arXiv (en), https://arxiv.org/abs/2609.35228 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_d47bb26e0d8d (Metraipolis archive dc_d47bb26e0d8d, sha256 5a73e5c82cb325acc9658d11bd1dc38fda6c4d09950c9ef4c2e498820283636a)
BibTeX
@misc{dcd47bb26e0d8d,
  title = {Token-Disentangled Latent Test-Time Scaling for Vision-Language Reasoning — arXiv:2609.35228},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2609.35228},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_d47bb26e0d8d; sha256 5a73e5c82cb325acc9658d11bd1dc38fda6c4d09950c9ef4c2e498820283636a},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2609.35228. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Latent test-time scaling improves reasoning by refining hidden states during inference, but existing methods typically apply a single scalar reward to all editable latent tokens. For multimodal large language models, this global update ignores that generated tokens play different roles: some are sensitive to visual evidence, while others correspond to uncertain reasoning decisions. We present Token-Disentangled Latent Test-Time Scaling, an inference-time framework that makes latent refinement token-role-aware. Starting from an initial generated trajectory, we optimize a short hidden-state prefix while routing perception-side visual feedback to image-sensitive tokens and reasoning feedback to high-entropy tokens. Tokens selected by neither route are constrained by an anchor regularizer. Across both perception and reasoning benchmarks on Qwen2.5-VL-7B and InternVL3.5-8B, our method lifts macro accuracy over CoT by +2.57 and +1.51 respectively, and outperforms strong output-space test-time scaling baselines under matched decoded-candidate budgets. Code is available at this https URL .

Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-09-29 in cs.AI).
Categorie arXiv: Computer Vision and Pattern Recognition (cs.CV) ; Artificial Intelligence (cs.AI).
Pagina del lavoro: https://arxiv.org/abs/2609.35228 — autori e testo completo sono lì (qui omessi i nomi: nessun dato personale nel deposito).
Licenza: titolo e abstract sono metadati arXiv distribuiti in CC0 1.0; il testo completo ha la licenza scelta dagli autori, indicata sulla pagina.
Letto il 2026-10-07T20:29:03.299Z dalla pagina https://arxiv.org/catchup/cs.AI/2026-09-29?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).