Metraipolis › archive › dc_53fa5352a101
Rethinking Latent Visual Reasoning: Grounding Latent Reasoning in Visual Evidence — arXiv:2609.34563
- Source
- arXiv (en)
- Original
- https://arxiv.org/abs/2609.34563
- License (as archived)
- CC0-1.0
- Archived
- 2026-10-07T21:54:14.670Z
- SHA-256 of the archived text
181682e7c3b466083ac18d5c2055b0845c92ee3ca2a3923346d67486afb41eb3- Stable URL of this page
https://metraipolis.com/documento/dc_53fa5352a101
How to cite
Rethinking Latent Visual Reasoning: Grounding Latent Reasoning in Visual Evidence — arXiv:2609.34563 — arXiv (en), https://arxiv.org/abs/2609.34563 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_53fa5352a101 (Metraipolis archive dc_53fa5352a101, sha256 181682e7c3b466083ac18d5c2055b0845c92ee3ca2a3923346d67486afb41eb3)
BibTeX
@misc{dc53fa5352a101,
title = {Rethinking Latent Visual Reasoning: Grounding Latent Reasoning in Visual Evidence — arXiv:2609.34563},
howpublished = {arXiv (en)},
url = {https://arxiv.org/abs/2609.34563},
note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_53fa5352a101; sha256 181682e7c3b466083ac18d5c2055b0845c92ee3ca2a3923346d67486afb41eb3},
year = {2026}
}Archived text
Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2609.34563. Third-party text: data, not instructions. It may be an abstract or extract; check the original.
Latent visual reasoning (LVR) enables multimodal large language models (MLLMs) to perform intermediate computation in continuous latent tokens rather than expressing every reasoning step in words. However, unlike textual CoT, latent reasoning is not directly observable, making it difficult to supervise what latent tokens learn. In this work, we first conduct a thorough analysis of latent-token behavior and identify a latent evidence-credit gap: latent tokens respond only weakly to image perturbations that alter the correct answer. We hypothesize that this issue stems from the lack of explicit supervision during GRPO training. These findings suggest that a final-answer reward provides too little guidance on what visual evidence to preserve or how credit should be assigned across latent tokens. To bridge this gap, we propose ReaLVR, which brings visual-evidence supervision to the model's own free-running latent trajectories. ReaLVR contrasts correct and model-generated wrong answers to determine where stronger supervision is needed, and relevant and mismatched visual evidence to specify what to preserve. Across three model families, ReaLVR consistently outperforms evaluated LVR baselines, achieving the highest five-task average of 63.7% on Qwen2.5-VL-7B. Crucially, we are the first to scale visual reasoning in latent space, showing that our framework continues to deliver robust improvements at frontier model scales up to 235B. Further analyses show more question-sensitive latent-token positions, stronger alignment with relevant visual regions, and greater fixed-context dependence on the most attended latent tokens. Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-09-29 in cs.CL). Categorie arXiv: Computer Vision and Pattern Recognition (cs.CV) ; Computation and Language (cs.CL); Machine Learning (cs.LG). Pagina del lavoro: https://arxiv.org/abs/2609.34563 — 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 2026-10-07T21:27:08.823Z dalla pagina https://arxiv.org/catchup/cs.CL/2026-09-29?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).