Saved in:
Bibliographic Details
Main Authors: Cui, Jin, Long, Xinyue, Zhang, Xunyong, Zhang, Yadong, Su, Chuanchang, Gan, Jingye, Zhao, Boran, Ren, Pengju
Format: Preprint
Published: 2026
Subjects:
Online Access:https://arxiv.org/abs/2605.07106
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866918489149669376
author Cui, Jin
Long, Xinyue
Zhang, Xunyong
Zhang, Yadong
Su, Chuanchang
Gan, Jingye
Zhao, Boran
Ren, Pengju
author_facet Cui, Jin
Long, Xinyue
Zhang, Xunyong
Zhang, Yadong
Su, Chuanchang
Gan, Jingye
Zhao, Boran
Ren, Pengju
contents Multimodal Large Language Models (MLLMs) have made remarkable progress on vision-language reasoning, yet most methods still compress visual evidence into discrete textual thoughts, creating an information bottleneck for fine-grained perception. Recent latent visual reasoning methods attempt to reason in continuous hidden states, but we find that they suffer from insufficient manifold compatibility: latent trajectories drift away from pretrained reasoning circuits, collapse into instance-agnostic patterns, and are often bypassed during answer generation. To address these issues, we propose RIS (Retrieve, Integrate, and Synthesize), a spatial-semantic grounded framework that develops latent reasoning as a compatible extension of pretrained MLLM computation. We first construct a step-wise grounded reasoning dataset with bounding boxes and region-specific semantic descriptions. Built on this supervision, RIS anchors latent tokens to both spatial and semantic evidence, enforces their causal role through a progressive attention bottleneck, and introduces short language transition tokens to bridge synthesized latent states back to vocabulary-aligned decoding. Experiments on V*, HRBench4K, HRBench8K, MMVP, and BLINK show consistent improvements over closed/open-source and latent reasoning baselines. Further analyses demonstrate that RIS learns diverse, interpretable, and progressively integrated latent trajectories, offering a practical path toward faithful internal visual reasoning in MLLMs.
format Preprint
id arxiv_https___arxiv_org_abs_2605_07106
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Retrieve, Integrate, and Synthesize: Spatial-Semantic Grounded Latent Visual Reasoning
Cui, Jin
Long, Xinyue
Zhang, Xunyong
Zhang, Yadong
Su, Chuanchang
Gan, Jingye
Zhao, Boran
Ren, Pengju
Computation and Language
Multimodal Large Language Models (MLLMs) have made remarkable progress on vision-language reasoning, yet most methods still compress visual evidence into discrete textual thoughts, creating an information bottleneck for fine-grained perception. Recent latent visual reasoning methods attempt to reason in continuous hidden states, but we find that they suffer from insufficient manifold compatibility: latent trajectories drift away from pretrained reasoning circuits, collapse into instance-agnostic patterns, and are often bypassed during answer generation. To address these issues, we propose RIS (Retrieve, Integrate, and Synthesize), a spatial-semantic grounded framework that develops latent reasoning as a compatible extension of pretrained MLLM computation. We first construct a step-wise grounded reasoning dataset with bounding boxes and region-specific semantic descriptions. Built on this supervision, RIS anchors latent tokens to both spatial and semantic evidence, enforces their causal role through a progressive attention bottleneck, and introduces short language transition tokens to bridge synthesized latent states back to vocabulary-aligned decoding. Experiments on V*, HRBench4K, HRBench8K, MMVP, and BLINK show consistent improvements over closed/open-source and latent reasoning baselines. Further analyses demonstrate that RIS learns diverse, interpretable, and progressively integrated latent trajectories, offering a practical path toward faithful internal visual reasoning in MLLMs.
title Retrieve, Integrate, and Synthesize: Spatial-Semantic Grounded Latent Visual Reasoning
topic Computation and Language
url https://arxiv.org/abs/2605.07106