<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="chen-etal-2026-two">
    <titleInfo>
        <title>Two to Tango: Coupled Task-Reference Selection for Safe LLM Fine-tuning</title>
    </titleInfo>
    <name type="personal">
        <namePart type="given">Xinrui</namePart>
        <namePart type="family">Chen</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Jianhao</namePart>
        <namePart type="family">Zhang</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Ou</namePart>
        <namePart type="family">Wu</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Di</namePart>
        <namePart type="family">Gao</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>Fine-tuning safety aligned large language models (LLMs) on downstream data improves adaptation but may erode learned safety behavior. Existing methods use fixed safety examples, global constraints, or one-sided task filtering. Our diagnostics show task updates expose different safety constraints, motivating joint selection of relevant references and compatible task samples. We propose DualSelect, a coupled framework for task and reference selection that refreshes task conditioned safety references before filtering whole task samples compatible with the induced reference direction. Under a minimax view, DualSelect selects safety references with high preservation loss and task conflict, together with compatible task samples, through entropy-regularized scoring surrogates, lazy reference refresh, and gradient correction. Across the evaluated English instruction-following and mathematical-reasoning settings on 1B–8B LLMs, DualSelect better preserves safety while maintaining task utility; using the REDORCA judge, it improves Safety Avg. over the strongest baseline by at least 5.10 points and remains highest in Safety Avg. across judges with moderate overhead. This view extends to retention focused continual learning. Code at: https://github.com/ChnIRuI/DualSelect.</abstract>
    <identifier type="citekey">chen-etal-2026-two</identifier>
    <location>
        <url>https://aclanthology.org/2026.emnlp-main.862/</url>
    </location>
    <part>
        <date>2026-10</date>
        <extent unit="page">
            <start>19944</start>
            <end>19968</end>
        </extent>
    </part>
</mods>
</modsCollection>

%0 Conference Proceedings
%T Two to Tango: Coupled Task-Reference Selection for Safe LLM Fine-tuning
%A Chen, Xinrui
%A Zhang, Jianhao
%A Wu, Ou
%A Gao, Di
%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 chen-etal-2026-two
%X Fine-tuning safety aligned large language models (LLMs) on downstream data improves adaptation but may erode learned safety behavior. Existing methods use fixed safety examples, global constraints, or one-sided task filtering. Our diagnostics show task updates expose different safety constraints, motivating joint selection of relevant references and compatible task samples. We propose DualSelect, a coupled framework for task and reference selection that refreshes task conditioned safety references before filtering whole task samples compatible with the induced reference direction. Under a minimax view, DualSelect selects safety references with high preservation loss and task conflict, together with compatible task samples, through entropy-regularized scoring surrogates, lazy reference refresh, and gradient correction. Across the evaluated English instruction-following and mathematical-reasoning settings on 1B–8B LLMs, DualSelect better preserves safety while maintaining task utility; using the REDORCA judge, it improves Safety Avg. over the strongest baseline by at least 5.10 points and remains highest in Safety Avg. across judges with moderate overhead. This view extends to retention focused continual learning. Code at: https://github.com/ChnIRuI/DualSelect.
%U https://aclanthology.org/2026.emnlp-main.862/
%P 19944-19968

Markdown (Informal)

[Two to Tango: Coupled Task-Reference Selection for Safe LLM Fine-tuning](https://aclanthology.org/2026.emnlp-main.862/) (Chen et al., EMNLP 2026)

• Two to Tango: Coupled Task-Reference Selection for Safe LLM Fine-tuning (Chen et al., EMNLP 2026)

ACL

• Xinrui Chen, Jianhao Zhang, Ou Wu, and Di Gao. 2026. Two to Tango: Coupled Task-Reference Selection for Safe LLM Fine-tuning. In Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing, pages 19944–19968, Budapest, Hungary. Association for Computational Linguistics.