<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="jian-etal-2026-learning">
    <titleInfo>
        <title>Learning When to Think in Latent Visual Space: Reasoning Mode Adaptive Reinforcement Learning</title>
    </titleInfo>
    <name type="personal">
        <namePart type="given">Paul</namePart>
        <namePart type="family">Jian</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Jinpeng</namePart>
        <namePart type="family">Ou</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Leo</namePart>
        <namePart type="family">Wei</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Qingpei</namePart>
        <namePart type="family">Guo</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Jianing</namePart>
        <namePart type="family">Li</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <originInfo>
        <dateIssued>2026-10</dateIssued>
    </originInfo>
    <typeOfResource>text</typeOfResource>
    <relatedItem type="host">
        <titleInfo>
            <title>Findings of the Association for Computational Linguistics: EMNLP 2026</title>
        </titleInfo>
        <name type="personal">
            <namePart type="given">Hung-Yi</namePart>
            <namePart type="family">Lee</namePart>
            <role>
                <roleTerm authority="marcrelator" type="text">editor</roleTerm>
            </role>
        </name>
        <name type="personal">
            <namePart type="given">Sunipa</namePart>
            <namePart type="family">Dev</namePart>
            <role>
                <roleTerm authority="marcrelator" type="text">editor</roleTerm>
            </role>
        </name>
        <name type="personal">
            <namePart type="given">Jessy</namePart>
            <namePart type="family">Li</namePart>
            <role>
                <roleTerm authority="marcrelator" type="text">editor</roleTerm>
            </role>
        </name>
        <name type="personal">
            <namePart type="given">Desmond</namePart>
            <namePart type="family">Elliott</namePart>
            <role>
                <roleTerm authority="marcrelator" type="text">editor</roleTerm>
            </role>
        </name>
        <originInfo>
            <publisher>Association for Computational Linguistics</publisher>
            <place>
                <placeTerm type="text">Budapest, Hungary</placeTerm>
            </place>
        </originInfo>
        <genre authority="marcgt">conference publication</genre>
        <identifier type="isbn">979-8-89176-462-0</identifier>
    </relatedItem>
    <abstract>Latent Visual Reasoning (LVR) unfolds reasoning in a continuous visual embedding space, markedly improving Vision-Language Models (VLMs) on vision-intensive tasks. However, our experiments reveal a hidden cost: LVR inevitably undermines VLMs’ strong textual reasoning ability, causing substantial degradation on logic-intensive tasks. This motivates us to unify both reasoning modes within a single VLM that adaptively selects the appropriate one. We propose MoLaT (Mixture of Latent and Textual Reasoning), a novel method that trains VLMs via reinforcement learning (RL) to select LVR and text-only reasoning adaptively. MoLaT is trained in two stages. Supervised fine-tuning endows the VLM with a dual-mode capability using mixed CoT data. Then, MoLaT introduces two complementary designs: 1) The mode-adaptive advantage decouples the mode-selection signal from per-rollout correctness reward at the token level: mode-selection tokens receive dedicated cross-mode gradients, while content tokens retain the group-relative advantage. 2) The entropy-modulated advantage weight uses sequence-level entropy as a difficulty proxy, modulating signal strength to prevent premature mode commitment. Experiments confirm that MoLaT improves both vision-intensive and logic-intensive abilities, validating adaptive mode selection as key to general-purpose reasoning VLMs.</abstract>
    <identifier type="citekey">jian-etal-2026-learning</identifier>
    <location>
        <url>https://aclanthology.org/2026.findings-emnlp.1986/</url>
    </location>
    <part>
        <date>2026-10</date>
        <extent unit="page">
            <start>42327</start>
            <end>42349</end>
        </extent>
    </part>
</mods>
</modsCollection>

%0 Conference Proceedings
%T Learning When to Think in Latent Visual Space: Reasoning Mode Adaptive Reinforcement Learning
%A Jian, Paul
%A Ou, Jinpeng
%A Wei, Leo
%A Guo, Qingpei
%A Li, Jianing
%Y Lee, Hung-Yi
%Y Dev, Sunipa
%Y Li, Jessy
%Y Elliott, Desmond
%S Findings of the Association for Computational Linguistics: EMNLP 2026
%D 2026
%8 October
%I Association for Computational Linguistics
%C Budapest, Hungary
%@ 979-8-89176-462-0
%F jian-etal-2026-learning
%X Latent Visual Reasoning (LVR) unfolds reasoning in a continuous visual embedding space, markedly improving Vision-Language Models (VLMs) on vision-intensive tasks. However, our experiments reveal a hidden cost: LVR inevitably undermines VLMs’ strong textual reasoning ability, causing substantial degradation on logic-intensive tasks. This motivates us to unify both reasoning modes within a single VLM that adaptively selects the appropriate one. We propose MoLaT (Mixture of Latent and Textual Reasoning), a novel method that trains VLMs via reinforcement learning (RL) to select LVR and text-only reasoning adaptively. MoLaT is trained in two stages. Supervised fine-tuning endows the VLM with a dual-mode capability using mixed CoT data. Then, MoLaT introduces two complementary designs: 1) The mode-adaptive advantage decouples the mode-selection signal from per-rollout correctness reward at the token level: mode-selection tokens receive dedicated cross-mode gradients, while content tokens retain the group-relative advantage. 2) The entropy-modulated advantage weight uses sequence-level entropy as a difficulty proxy, modulating signal strength to prevent premature mode commitment. Experiments confirm that MoLaT improves both vision-intensive and logic-intensive abilities, validating adaptive mode selection as key to general-purpose reasoning VLMs.
%U https://aclanthology.org/2026.findings-emnlp.1986/
%P 42327-42349

Markdown (Informal)

[Learning When to Think in Latent Visual Space: Reasoning Mode Adaptive Reinforcement Learning](https://aclanthology.org/2026.findings-emnlp.1986/) (Jian et al., Findings 2026)

• Learning When to Think in Latent Visual Space: Reasoning Mode Adaptive Reinforcement Learning (Jian et al., Findings 2026)

ACL

• Paul Jian, Jinpeng Ou, Leo Wei, Qingpei Guo, and Jianing Li. 2026. Learning When to Think in Latent Visual Space: Reasoning Mode Adaptive Reinforcement Learning. In Findings of the Association for Computational Linguistics: EMNLP 2026, pages 42327–42349, Budapest, Hungary. Association for Computational Linguistics.