<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="wedgwood-etal-2026-dspa">
    <titleInfo>
        <title>DSPA: Dynamic SAE Steering for Data-Efficient Preference Alignment</title>
    </titleInfo>
    <name type="personal">
        <namePart type="given">James</namePart>
        <namePart type="family">Wedgwood</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Aashiq</namePart>
        <namePart type="family">Muhamed</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Mona</namePart>
        <namePart type="given">T</namePart>
        <namePart type="family">Diab</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Virginia</namePart>
        <namePart type="family">Smith</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>Preference alignment is usually achieved by weight-updating training on preference data, which is costly and provides limited mechanistic visibility. We propose Dynamic SAE Steering for Preference Alignment (DSPA), an inference-time method that makes sparse autoencoder (SAE) steering prompt-conditional. From preference triples, DSPA computes a conditional-difference map linking prompt features to generation-control features; during decoding, it modifies only token-active latents, without base-model weight updates. Across Gemma-2-2B/9B and Qwen3-8B, DSPA improves MT-Bench and is competitive on AlpacaEval while preserving multiple-choice accuracy. Under restricted preference data, DSPA remains robust and can rival heavier alignment pipelines at a fraction of the compute. Finally, we audit the SAE features DSPA modifies, finding that preference directions are dominated by discourse and stylistic signals, and provide theory clarifying the conditional-difference map estimate and when top-k ablation is principled.</abstract>
    <identifier type="citekey">wedgwood-etal-2026-dspa</identifier>
    <location>
        <url>https://aclanthology.org/2026.emnlp-main.169/</url>
    </location>
    <part>
        <date>2026-10</date>
        <extent unit="page">
            <start>3879</start>
            <end>3898</end>
        </extent>
    </part>
</mods>
</modsCollection>

%0 Conference Proceedings
%T DSPA: Dynamic SAE Steering for Data-Efficient Preference Alignment
%A Wedgwood, James
%A Muhamed, Aashiq
%A Diab, Mona T.
%A Smith, Virginia
%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 wedgwood-etal-2026-dspa
%X Preference alignment is usually achieved by weight-updating training on preference data, which is costly and provides limited mechanistic visibility. We propose Dynamic SAE Steering for Preference Alignment (DSPA), an inference-time method that makes sparse autoencoder (SAE) steering prompt-conditional. From preference triples, DSPA computes a conditional-difference map linking prompt features to generation-control features; during decoding, it modifies only token-active latents, without base-model weight updates. Across Gemma-2-2B/9B and Qwen3-8B, DSPA improves MT-Bench and is competitive on AlpacaEval while preserving multiple-choice accuracy. Under restricted preference data, DSPA remains robust and can rival heavier alignment pipelines at a fraction of the compute. Finally, we audit the SAE features DSPA modifies, finding that preference directions are dominated by discourse and stylistic signals, and provide theory clarifying the conditional-difference map estimate and when top-k ablation is principled.
%U https://aclanthology.org/2026.emnlp-main.169/
%P 3879-3898

Markdown (Informal)

[DSPA: Dynamic SAE Steering for Data-Efficient Preference Alignment](https://aclanthology.org/2026.emnlp-main.169/) (Wedgwood et al., EMNLP 2026)

• DSPA: Dynamic SAE Steering for Data-Efficient Preference Alignment (Wedgwood et al., EMNLP 2026)

ACL

• James Wedgwood, Aashiq Muhamed, Mona T. Diab, and Virginia Smith. 2026. DSPA: Dynamic SAE Steering for Data-Efficient Preference Alignment. In Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing, pages 3879–3898, Budapest, Hungary. Association for Computational Linguistics.