<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="prasanth-2026-efficient">
    <titleInfo>
        <title>Efficient Visual Grounding in VQA via Question-Guided Sparse Attention</title>
    </titleInfo>
    <name>
        <namePart>Prasanth</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>Proceedings of the 4th Workshop on Advances in Language and Vision Research (ALVR)</title>
        </titleInfo>
        <name type="personal">
            <namePart type="given">Qianqi</namePart>
            <namePart type="family">Yan</namePart>
            <role>
                <roleTerm authority="marcrelator" type="text">editor</roleTerm>
            </role>
        </name>
        <name type="personal">
            <namePart type="given">Syrielle</namePart>
            <namePart type="family">Montariol</namePart>
            <role>
                <roleTerm authority="marcrelator" type="text">editor</roleTerm>
            </role>
        </name>
        <name type="personal">
            <namePart type="given">Yue</namePart>
            <namePart type="family">Fan</namePart>
            <role>
                <roleTerm authority="marcrelator" type="text">editor</roleTerm>
            </role>
        </name>
        <name type="personal">
            <namePart type="given">Jing</namePart>
            <namePart type="family">Gu</namePart>
            <role>
                <roleTerm authority="marcrelator" type="text">editor</roleTerm>
            </role>
        </name>
        <name type="personal">
            <namePart type="given">Jiayi</namePart>
            <namePart type="family">Pan</namePart>
            <role>
                <roleTerm authority="marcrelator" type="text">editor</roleTerm>
            </role>
        </name>
        <name type="personal">
            <namePart type="given">Manling</namePart>
            <namePart type="family">Li</namePart>
            <role>
                <roleTerm authority="marcrelator" type="text">editor</roleTerm>
            </role>
        </name>
        <name type="personal">
            <namePart type="given">Parisa</namePart>
            <namePart type="family">Kordjamshidi</namePart>
            <role>
                <roleTerm authority="marcrelator" type="text">editor</roleTerm>
            </role>
        </name>
        <name type="personal">
            <namePart type="given">Alane</namePart>
            <namePart type="family">Suhr</namePart>
            <role>
                <roleTerm authority="marcrelator" type="text">editor</roleTerm>
            </role>
        </name>
        <name type="personal">
            <namePart type="given">Xin</namePart>
            <namePart type="given">Eric</namePart>
            <namePart type="family">Wang</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, USA</placeTerm>
            </place>
        </originInfo>
        <genre authority="marcgt">conference publication</genre>
        <identifier type="isbn">979-8-89176-398-2</identifier>
    </relatedItem>
    <abstract>Visual Question Answering (VQA) models process all image patches uniformlydespite questions typically requiring only a small subset of visual information.This inefficiency leads to unnecessary computation and can result in attentiondilution across irrelevant image regions. We propose Question-GuidedSparse Attention (QGSA), a plug-and-play mechanism that dynamically selectsrelevant image patches conditioned on question semantics. Our approach introducesthree components: (1)a differentiable patch selector based on Gumbel-Softmaxreparameterisation that enables end-to-end training with hard patch selection atinference; (2)a self-supervised grounding loss that encourages spatialselectivity without bounding-box annotations, combining contrastive patchselection with patch–word alignment via a frozen CLIP encoder; and (3)anadaptive sparsity mechanism that adjusts the number of selected patches accordingto estimated question complexity. Experiments on SmolVLM-256M-Instruct andSmolVLM-500M-Instruct across three VQA benchmarks (VQA-RAD, A-OKVQA, RefCOCO)demonstrate that QGSA reduces cross-attention FLOPs by 91–99% across inputresolutions, achieving up to 76\times theoretical speedup at 576px resolution, whilemaintaining exact accuracy parity with the dense baseline (Δ=0.0 ppon all datasets).Wall-clock parity with the dense baseline is reached at 336px; realisedend-to-end speedup requires larger models where cross-attention dominates totalcompute. QGSA consistently selects an average of k\approx17 patches out of576 (256M model), up to k\approx18 (500M model), yielding up to a 34\timesreduction in the visual token sequence. These small-scale results validate thefeasibility of question-conditioned sparse attention and provide a foundation forscaling to larger VLMs.</abstract>
    <identifier type="citekey">prasanth-2026-efficient</identifier>
    <location>
        <url>https://aclanthology.org/2026.alvr-main.24/</url>
    </location>
    <part>
        <date>2026-07</date>
        <extent unit="page">
            <start>260</start>
            <end>271</end>
        </extent>
    </part>
</mods>
</modsCollection>

%0 Conference Proceedings
%T Efficient Visual Grounding in VQA via Question-Guided Sparse Attention
%Y Yan, Qianqi
%Y Montariol, Syrielle
%Y Fan, Yue
%Y Gu, Jing
%Y Pan, Jiayi
%Y Li, Manling
%Y Kordjamshidi, Parisa
%Y Suhr, Alane
%Y Wang, Xin Eric
%A Prasanth
%S Proceedings of the 4th Workshop on Advances in Language and Vision Research (ALVR)
%D 2026
%8 July
%I Association for Computational Linguistics
%C San Diego, California, USA
%@ 979-8-89176-398-2
%F prasanth-2026-efficient
%X Visual Question Answering (VQA) models process all image patches uniformlydespite questions typically requiring only a small subset of visual information.This inefficiency leads to unnecessary computation and can result in attentiondilution across irrelevant image regions. We propose Question-GuidedSparse Attention (QGSA), a plug-and-play mechanism that dynamically selectsrelevant image patches conditioned on question semantics. Our approach introducesthree components: (1)a differentiable patch selector based on Gumbel-Softmaxreparameterisation that enables end-to-end training with hard patch selection atinference; (2)a self-supervised grounding loss that encourages spatialselectivity without bounding-box annotations, combining contrastive patchselection with patch–word alignment via a frozen CLIP encoder; and (3)anadaptive sparsity mechanism that adjusts the number of selected patches accordingto estimated question complexity. Experiments on SmolVLM-256M-Instruct andSmolVLM-500M-Instruct across three VQA benchmarks (VQA-RAD, A-OKVQA, RefCOCO)demonstrate that QGSA reduces cross-attention FLOPs by 91–99% across inputresolutions, achieving up to 76\times theoretical speedup at 576px resolution, whilemaintaining exact accuracy parity with the dense baseline (Δ=0.0 ppon all datasets).Wall-clock parity with the dense baseline is reached at 336px; realisedend-to-end speedup requires larger models where cross-attention dominates totalcompute. QGSA consistently selects an average of k\approx17 patches out of576 (256M model), up to k\approx18 (500M model), yielding up to a 34\timesreduction in the visual token sequence. These small-scale results validate thefeasibility of question-conditioned sparse attention and provide a foundation forscaling to larger VLMs.
%U https://aclanthology.org/2026.alvr-main.24/
%P 260-271

Markdown (Informal)

[Efficient Visual Grounding in VQA via Question-Guided Sparse Attention](https://aclanthology.org/2026.alvr-main.24/) (Prasanth, ALVR 2026)

• Efficient Visual Grounding in VQA via Question-Guided Sparse Attention (Prasanth, ALVR 2026)

ACL

• Prasanth. 2026. Efficient Visual Grounding in VQA via Question-Guided Sparse Attention. In Proceedings of the 4th Workshop on Advances in Language and Vision Research (ALVR), pages 260–271, San Diego, California, USA. Association for Computational Linguistics.