@inproceedings{li-etal-2026-role-entropy,
title = "The Role of Entropy in Visual Grounding: Analysis and Optimization",
author = "Li, Shuo and
Sun, Jiajun and
Zhang, Zhihao and
Fan, Xiaoran and
Jin, Senjie and
Li, Hui and
Yang, Yuming and
Ye, Junjie and
Shen, Lixing and
Ji, Tao and
Gui, Tao and
Zhang, Qi and
Huang, Xuanjing",
editor = "Lee, Hung-Yi and
Dev, Sunipa and
Li, Jessy and
Elliott, Desmond",
booktitle = "Findings of the {A}ssociation for {C}omputational {L}inguistics: {EMNLP} 2026",
month = oct,
year = "2026",
address = "Budapest, Hungary",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.findings-emnlp.1296/",
pages = "27406--27427",
ISBN = "979-8-89176-462-0",
abstract = "Recent advances in reinforcement learning{--}based fine-tuning of multimodal large language models (MLLMs) have highlighted the importance of entropy control for stable optimization. While entropy has been extensively studied in reasoning-centric settings, its behavior and optimization role in perception-oriented tasks remain insufficiently understood. In this work, we reveal that entropy exhibits fundamentally different roles in perception tasks compared to reasoning, particularly under continuous prediction and reward. Using visual grounding as a case study, we show that entropy can remain misaligned with optimization objectives due to persistent low-probability yet high-advantage samples. Building on these findings, we introduce ECVGPO (Entropy Control Visual Grounding Policy Optimization), an interpretable algorithm designed for effective entropy regulation. Through entropy control, the trade-off between exploration and exploitation is better balanced. Experiments show that ECVGPO achieves broad improvements across various benchmarks and models."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="li-etal-2026-role-entropy">
<titleInfo>
<title>The Role of Entropy in Visual Grounding: Analysis and Optimization</title>
</titleInfo>
<name type="personal">
<namePart type="given">Shuo</namePart>
<namePart type="family">Li</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jiajun</namePart>
<namePart type="family">Sun</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Zhihao</namePart>
<namePart type="family">Zhang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Xiaoran</namePart>
<namePart type="family">Fan</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Senjie</namePart>
<namePart type="family">Jin</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Hui</namePart>
<namePart type="family">Li</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yuming</namePart>
<namePart type="family">Yang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Junjie</namePart>
<namePart type="family">Ye</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Lixing</namePart>
<namePart type="family">Shen</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Tao</namePart>
<namePart type="family">Ji</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Tao</namePart>
<namePart type="family">Gui</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Qi</namePart>
<namePart type="family">Zhang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Xuanjing</namePart>
<namePart type="family">Huang</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>Recent advances in reinforcement learning–based fine-tuning of multimodal large language models (MLLMs) have highlighted the importance of entropy control for stable optimization. While entropy has been extensively studied in reasoning-centric settings, its behavior and optimization role in perception-oriented tasks remain insufficiently understood. In this work, we reveal that entropy exhibits fundamentally different roles in perception tasks compared to reasoning, particularly under continuous prediction and reward. Using visual grounding as a case study, we show that entropy can remain misaligned with optimization objectives due to persistent low-probability yet high-advantage samples. Building on these findings, we introduce ECVGPO (Entropy Control Visual Grounding Policy Optimization), an interpretable algorithm designed for effective entropy regulation. Through entropy control, the trade-off between exploration and exploitation is better balanced. Experiments show that ECVGPO achieves broad improvements across various benchmarks and models.</abstract>
<identifier type="citekey">li-etal-2026-role-entropy</identifier>
<location>
<url>https://aclanthology.org/2026.findings-emnlp.1296/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>27406</start>
<end>27427</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T The Role of Entropy in Visual Grounding: Analysis and Optimization
%A Li, Shuo
%A Sun, Jiajun
%A Zhang, Zhihao
%A Fan, Xiaoran
%A Jin, Senjie
%A Li, Hui
%A Yang, Yuming
%A Ye, Junjie
%A Shen, Lixing
%A Ji, Tao
%A Gui, Tao
%A Zhang, Qi
%A Huang, Xuanjing
%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 li-etal-2026-role-entropy
%X Recent advances in reinforcement learning–based fine-tuning of multimodal large language models (MLLMs) have highlighted the importance of entropy control for stable optimization. While entropy has been extensively studied in reasoning-centric settings, its behavior and optimization role in perception-oriented tasks remain insufficiently understood. In this work, we reveal that entropy exhibits fundamentally different roles in perception tasks compared to reasoning, particularly under continuous prediction and reward. Using visual grounding as a case study, we show that entropy can remain misaligned with optimization objectives due to persistent low-probability yet high-advantage samples. Building on these findings, we introduce ECVGPO (Entropy Control Visual Grounding Policy Optimization), an interpretable algorithm designed for effective entropy regulation. Through entropy control, the trade-off between exploration and exploitation is better balanced. Experiments show that ECVGPO achieves broad improvements across various benchmarks and models.
%U https://aclanthology.org/2026.findings-emnlp.1296/
%P 27406-27427
Markdown (Informal)
[The Role of Entropy in Visual Grounding: Analysis and Optimization](https://aclanthology.org/2026.findings-emnlp.1296/) (Li et al., Findings 2026)
ACL
- Shuo Li, Jiajun Sun, Zhihao Zhang, Xiaoran Fan, Senjie Jin, Hui Li, Yuming Yang, Junjie Ye, Lixing Shen, Tao Ji, Tao Gui, Qi Zhang, and Xuanjing Huang. 2026. The Role of Entropy in Visual Grounding: Analysis and Optimization. In Findings of the Association for Computational Linguistics: EMNLP 2026, pages 27406–27427, Budapest, Hungary. Association for Computational Linguistics.