@inproceedings{fime-etal-2026-lvlms,
title = "Do {LVLM}s Know When They Are Wrong? Trajectory-Based Correctness Estimation from a Single Forward Pass",
author = "Fime, Awal Ahmed and
Hossain, Md Zarif and
Imteaj, Ahmed",
editor = "Lee, Hung-Yi and
Dev, Sunipa and
Li, Jessy and
Elliott, Desmond",
booktitle = "Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing",
month = oct,
year = "2026",
address = "Budapest, Hungary",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.emnlp-main.2468/",
pages = "56033--56050",
ISBN = "979-8-89176-459-0",
abstract = "Large vision-language models (LVLMs) exhibit strong multimodal reasoning capabilities, yet reliably estimating whether a generated response is correct remains a fundamental challenge. Existing confidence estimation methods commonly rely on repeated querying, self-consistency sampling, or layer-wise probing, which introduce additional inference overhead, self-preference bias, or costly full-depth analysis. In this work, we propose a trajectory-based framework for estimating LVLM response correctness from a single forward generation pass. Instead of treating internal representations as isolated layer-wise signals, we model response generation as a structured trajectory through the LVLM, capturing the evolution of visual evidence in the vision encoder, geometric transformations induced by cross-modal projection, and probability dynamics within the language decoder. Our key insight is that response correctness is encoded in the geometry and dynamics of these component-specific trajectories. Based on this observation, we derive lightweight trajectory descriptors for each major LVLM module and integrate them through a calibrated correctness estimator trained with as few as 500 labeled examples. At test time, our method requires no additional inference passes and no ground-truth labels. Extensive experiments across multiple LVLM architectures and benchmark datasets demonstrate consistent improvements over existing baselines, including up to 16.44 higher AUROC on Qwen-2B over VisWiz and 15.61 higher AUPR on LLaVA-Next-7B over ScienceQA compared to the strongest competing methods."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="fime-etal-2026-lvlms">
<titleInfo>
<title>Do LVLMs Know When They Are Wrong? Trajectory-Based Correctness Estimation from a Single Forward Pass</title>
</titleInfo>
<name type="personal">
<namePart type="given">Awal</namePart>
<namePart type="given">Ahmed</namePart>
<namePart type="family">Fime</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Md</namePart>
<namePart type="given">Zarif</namePart>
<namePart type="family">Hossain</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ahmed</namePart>
<namePart type="family">Imteaj</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>Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing</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">979-8-89176-459-0</identifier>
</relatedItem>
<abstract>Large vision-language models (LVLMs) exhibit strong multimodal reasoning capabilities, yet reliably estimating whether a generated response is correct remains a fundamental challenge. Existing confidence estimation methods commonly rely on repeated querying, self-consistency sampling, or layer-wise probing, which introduce additional inference overhead, self-preference bias, or costly full-depth analysis. In this work, we propose a trajectory-based framework for estimating LVLM response correctness from a single forward generation pass. Instead of treating internal representations as isolated layer-wise signals, we model response generation as a structured trajectory through the LVLM, capturing the evolution of visual evidence in the vision encoder, geometric transformations induced by cross-modal projection, and probability dynamics within the language decoder. Our key insight is that response correctness is encoded in the geometry and dynamics of these component-specific trajectories. Based on this observation, we derive lightweight trajectory descriptors for each major LVLM module and integrate them through a calibrated correctness estimator trained with as few as 500 labeled examples. At test time, our method requires no additional inference passes and no ground-truth labels. Extensive experiments across multiple LVLM architectures and benchmark datasets demonstrate consistent improvements over existing baselines, including up to 16.44 higher AUROC on Qwen-2B over VisWiz and 15.61 higher AUPR on LLaVA-Next-7B over ScienceQA compared to the strongest competing methods.</abstract>
<identifier type="citekey">fime-etal-2026-lvlms</identifier>
<location>
<url>https://aclanthology.org/2026.emnlp-main.2468/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>56033</start>
<end>56050</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Do LVLMs Know When They Are Wrong? Trajectory-Based Correctness Estimation from a Single Forward Pass
%A Fime, Awal Ahmed
%A Hossain, Md Zarif
%A Imteaj, Ahmed
%Y Lee, Hung-Yi
%Y Dev, Sunipa
%Y Li, Jessy
%Y Elliott, Desmond
%S Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing
%D 2026
%8 October
%I Association for Computational Linguistics
%C Budapest, Hungary
%@ 979-8-89176-459-0
%F fime-etal-2026-lvlms
%X Large vision-language models (LVLMs) exhibit strong multimodal reasoning capabilities, yet reliably estimating whether a generated response is correct remains a fundamental challenge. Existing confidence estimation methods commonly rely on repeated querying, self-consistency sampling, or layer-wise probing, which introduce additional inference overhead, self-preference bias, or costly full-depth analysis. In this work, we propose a trajectory-based framework for estimating LVLM response correctness from a single forward generation pass. Instead of treating internal representations as isolated layer-wise signals, we model response generation as a structured trajectory through the LVLM, capturing the evolution of visual evidence in the vision encoder, geometric transformations induced by cross-modal projection, and probability dynamics within the language decoder. Our key insight is that response correctness is encoded in the geometry and dynamics of these component-specific trajectories. Based on this observation, we derive lightweight trajectory descriptors for each major LVLM module and integrate them through a calibrated correctness estimator trained with as few as 500 labeled examples. At test time, our method requires no additional inference passes and no ground-truth labels. Extensive experiments across multiple LVLM architectures and benchmark datasets demonstrate consistent improvements over existing baselines, including up to 16.44 higher AUROC on Qwen-2B over VisWiz and 15.61 higher AUPR on LLaVA-Next-7B over ScienceQA compared to the strongest competing methods.
%U https://aclanthology.org/2026.emnlp-main.2468/
%P 56033-56050
Markdown (Informal)
[Do LVLMs Know When They Are Wrong? Trajectory-Based Correctness Estimation from a Single Forward Pass](https://aclanthology.org/2026.emnlp-main.2468/) (Fime et al., EMNLP 2026)
ACL