@inproceedings{sunil-etal-2026-ireasoner,
title = "{IREASONER}: Trajectory-Aware Intrinsic Reasoning Supervision for Self-Evolving Large Multimodal Models",
author = "Sunil, Meghana and
Venmathimaran, Manikandarajan and
Kavitha, Muthu Subash",
editor = "Liakata, Maria and
Moreira, Viviane P. and
Zhang, Jiajun and
Jurgens, David",
booktitle = "Findings of the {A}ssociation for {C}omputational {L}inguistics: {ACL} 2026",
month = jul,
year = "2026",
address = "San Diego, California, United States",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.findings-acl.1468/",
doi = "10.18653/v1/2026.findings-acl.1468",
pages = "29361--29372",
ISBN = "979-8-89176-395-1",
abstract = "Recent work shows that large multimodal models (LMMs) can self-improve from unlabeled data via self-play and intrinsic feedback. Yet existing self-evolving frameworks mainly reward final outcomes, leaving intermediate reasoning weakly constrained despite its importance for visually grounded decision making. We propose IREASONER, a self-evolving framework that improves an LMM{'}s implicit reasoning by explicitly eliciting chain-of-thought (CoT) and rewarding its internal agreement. In a Proposer{--}Solver loop over unlabeled images, IREASONER augments outcome-level intrinsic rewards with a trajectory-aware signal defined over intermediate reasoning steps, providing learning signals that distinguish reasoning paths leading to the same answer without ground-truth labels or external judges. Starting from Qwen2.5-VL-7B, IREASONER yields up to +2.1 points across diverse multimodal reasoning benchmarks under fully unsupervised post-training. We hope this work serves as a starting point for reasoning-aware self-improvement in LMMs in purely unsupervised settings."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="sunil-etal-2026-ireasoner">
<titleInfo>
<title>IREASONER: Trajectory-Aware Intrinsic Reasoning Supervision for Self-Evolving Large Multimodal Models</title>
</titleInfo>
<name type="personal">
<namePart type="given">Meghana</namePart>
<namePart type="family">Sunil</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Manikandarajan</namePart>
<namePart type="family">Venmathimaran</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Muthu</namePart>
<namePart type="given">Subash</namePart>
<namePart type="family">Kavitha</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2026-07</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Findings of the Association for Computational Linguistics: ACL 2026</title>
</titleInfo>
<name type="personal">
<namePart type="given">Maria</namePart>
<namePart type="family">Liakata</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Viviane</namePart>
<namePart type="given">P</namePart>
<namePart type="family">Moreira</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jiajun</namePart>
<namePart type="family">Zhang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">David</namePart>
<namePart type="family">Jurgens</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">San Diego, California, United States</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
<identifier type="isbn">979-8-89176-395-1</identifier>
</relatedItem>
<abstract>Recent work shows that large multimodal models (LMMs) can self-improve from unlabeled data via self-play and intrinsic feedback. Yet existing self-evolving frameworks mainly reward final outcomes, leaving intermediate reasoning weakly constrained despite its importance for visually grounded decision making. We propose IREASONER, a self-evolving framework that improves an LMM’s implicit reasoning by explicitly eliciting chain-of-thought (CoT) and rewarding its internal agreement. In a Proposer–Solver loop over unlabeled images, IREASONER augments outcome-level intrinsic rewards with a trajectory-aware signal defined over intermediate reasoning steps, providing learning signals that distinguish reasoning paths leading to the same answer without ground-truth labels or external judges. Starting from Qwen2.5-VL-7B, IREASONER yields up to +2.1 points across diverse multimodal reasoning benchmarks under fully unsupervised post-training. We hope this work serves as a starting point for reasoning-aware self-improvement in LMMs in purely unsupervised settings.</abstract>
<identifier type="citekey">sunil-etal-2026-ireasoner</identifier>
<identifier type="doi">10.18653/v1/2026.findings-acl.1468</identifier>
<location>
<url>https://aclanthology.org/2026.findings-acl.1468/</url>
</location>
<part>
<date>2026-07</date>
<extent unit="page">
<start>29361</start>
<end>29372</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T IREASONER: Trajectory-Aware Intrinsic Reasoning Supervision for Self-Evolving Large Multimodal Models
%A Sunil, Meghana
%A Venmathimaran, Manikandarajan
%A Kavitha, Muthu Subash
%Y Liakata, Maria
%Y Moreira, Viviane P.
%Y Zhang, Jiajun
%Y Jurgens, David
%S Findings of the Association for Computational Linguistics: ACL 2026
%D 2026
%8 July
%I Association for Computational Linguistics
%C San Diego, California, United States
%@ 979-8-89176-395-1
%F sunil-etal-2026-ireasoner
%X Recent work shows that large multimodal models (LMMs) can self-improve from unlabeled data via self-play and intrinsic feedback. Yet existing self-evolving frameworks mainly reward final outcomes, leaving intermediate reasoning weakly constrained despite its importance for visually grounded decision making. We propose IREASONER, a self-evolving framework that improves an LMM’s implicit reasoning by explicitly eliciting chain-of-thought (CoT) and rewarding its internal agreement. In a Proposer–Solver loop over unlabeled images, IREASONER augments outcome-level intrinsic rewards with a trajectory-aware signal defined over intermediate reasoning steps, providing learning signals that distinguish reasoning paths leading to the same answer without ground-truth labels or external judges. Starting from Qwen2.5-VL-7B, IREASONER yields up to +2.1 points across diverse multimodal reasoning benchmarks under fully unsupervised post-training. We hope this work serves as a starting point for reasoning-aware self-improvement in LMMs in purely unsupervised settings.
%R 10.18653/v1/2026.findings-acl.1468
%U https://aclanthology.org/2026.findings-acl.1468/
%U https://doi.org/10.18653/v1/2026.findings-acl.1468
%P 29361-29372
Markdown (Informal)
[IREASONER: Trajectory-Aware Intrinsic Reasoning Supervision for Self-Evolving Large Multimodal Models](https://aclanthology.org/2026.findings-acl.1468/) (Sunil et al., Findings 2026)
ACL