@inproceedings{li-etal-2026-use,
title = "When and How to use In-context Learning for modern {LVLM}s? An Extensive Investigation",
author = "Li, Pengxiang and
Zhao, Yu and
Yu, Jiaxin and
Xie, Liang and
Yin, Erwei",
editor = "Zhang, Yue and
Xu, Ruifeng and
Purwarianti, Ayu and
Rong, Tong and
Suzuki, Jun",
booktitle = "Findings of the {A}ssociation for {C}omputational {L}inguistics: {IJCNLP}-{AACL} 2026",
month = nov,
year = "2026",
address = "Hengqin, China",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.findings-aacl.162/",
pages = "3228--3249",
ISBN = "979-8-89176-444-6",
abstract = "Multi-Modal In-Context Learning (MM-ICL) is commonly employed to enhance performance of Large Vision-Language Models (LVLMs) on downstream tasks. However, as next-generation models like QwenVL and InternVL have been optimized through instruction tuning and alignment, their baseline capabilities have been strengthened. This necessitates a re-evaluation of the effectiveness and optimal configuration of MM-ICL across different models. By evaluating diverse LVLMs spanning different parameter scales (7B{--}32B) and accessibility settings (open-source and closed-source) across eight cross-domain datasets, this paper systematically investigates the performance of MM-ICL under varying model capabilities and task settings. Experimental results indicate that, beyond validating established patterns, next-generation models exhibit distinct phenomena: in certain logical reasoning and format-sensitive tasks, the introduction of MM-ICL does not necessarily yield improvements and can even lead to performance degradation in specific scenarios. We analyze these observations and, based on extensive experimental data, provide new recommendations for selecting MM-ICL strategies for LVLMs."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="li-etal-2026-use">
<titleInfo>
<title>When and How to use In-context Learning for modern LVLMs? An Extensive Investigation</title>
</titleInfo>
<name type="personal">
<namePart type="given">Pengxiang</namePart>
<namePart type="family">Li</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yu</namePart>
<namePart type="family">Zhao</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jiaxin</namePart>
<namePart type="family">Yu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Liang</namePart>
<namePart type="family">Xie</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Erwei</namePart>
<namePart type="family">Yin</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2026-11</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Findings of the Association for Computational Linguistics: IJCNLP-AACL 2026</title>
</titleInfo>
<name type="personal">
<namePart type="given">Yue</namePart>
<namePart type="family">Zhang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ruifeng</namePart>
<namePart type="family">Xu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ayu</namePart>
<namePart type="family">Purwarianti</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Tong</namePart>
<namePart type="family">Rong</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jun</namePart>
<namePart type="family">Suzuki</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">Hengqin, China</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
<identifier type="isbn">979-8-89176-444-6</identifier>
</relatedItem>
<abstract>Multi-Modal In-Context Learning (MM-ICL) is commonly employed to enhance performance of Large Vision-Language Models (LVLMs) on downstream tasks. However, as next-generation models like QwenVL and InternVL have been optimized through instruction tuning and alignment, their baseline capabilities have been strengthened. This necessitates a re-evaluation of the effectiveness and optimal configuration of MM-ICL across different models. By evaluating diverse LVLMs spanning different parameter scales (7B–32B) and accessibility settings (open-source and closed-source) across eight cross-domain datasets, this paper systematically investigates the performance of MM-ICL under varying model capabilities and task settings. Experimental results indicate that, beyond validating established patterns, next-generation models exhibit distinct phenomena: in certain logical reasoning and format-sensitive tasks, the introduction of MM-ICL does not necessarily yield improvements and can even lead to performance degradation in specific scenarios. We analyze these observations and, based on extensive experimental data, provide new recommendations for selecting MM-ICL strategies for LVLMs.</abstract>
<identifier type="citekey">li-etal-2026-use</identifier>
<location>
<url>https://aclanthology.org/2026.findings-aacl.162/</url>
</location>
<part>
<date>2026-11</date>
<extent unit="page">
<start>3228</start>
<end>3249</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T When and How to use In-context Learning for modern LVLMs? An Extensive Investigation
%A Li, Pengxiang
%A Zhao, Yu
%A Yu, Jiaxin
%A Xie, Liang
%A Yin, Erwei
%Y Zhang, Yue
%Y Xu, Ruifeng
%Y Purwarianti, Ayu
%Y Rong, Tong
%Y Suzuki, Jun
%S Findings of the Association for Computational Linguistics: IJCNLP-AACL 2026
%D 2026
%8 November
%I Association for Computational Linguistics
%C Hengqin, China
%@ 979-8-89176-444-6
%F li-etal-2026-use
%X Multi-Modal In-Context Learning (MM-ICL) is commonly employed to enhance performance of Large Vision-Language Models (LVLMs) on downstream tasks. However, as next-generation models like QwenVL and InternVL have been optimized through instruction tuning and alignment, their baseline capabilities have been strengthened. This necessitates a re-evaluation of the effectiveness and optimal configuration of MM-ICL across different models. By evaluating diverse LVLMs spanning different parameter scales (7B–32B) and accessibility settings (open-source and closed-source) across eight cross-domain datasets, this paper systematically investigates the performance of MM-ICL under varying model capabilities and task settings. Experimental results indicate that, beyond validating established patterns, next-generation models exhibit distinct phenomena: in certain logical reasoning and format-sensitive tasks, the introduction of MM-ICL does not necessarily yield improvements and can even lead to performance degradation in specific scenarios. We analyze these observations and, based on extensive experimental data, provide new recommendations for selecting MM-ICL strategies for LVLMs.
%U https://aclanthology.org/2026.findings-aacl.162/
%P 3228-3249
Markdown (Informal)
[When and How to use In-context Learning for modern LVLMs? An Extensive Investigation](https://aclanthology.org/2026.findings-aacl.162/) (Li et al., Findings 2026)
ACL