Denoising Concept Vectors with Sparse Autoencoders for Improved Language Model Steering

Haiyan Zhao; Xuansheng Wu; Fan Yang; Bo Shen; Ninghao Liu; Mengnan Du

Denoising Concept Vectors with Sparse Autoencoders for Improved Language Model Steering

Haiyan Zhao, Xuansheng Wu, Fan Yang, Bo Shen, Ninghao Liu, Mengnan Du

Correct Metadata for

Use this form to create a GitHub issue with structured data describing the correction. You will need a GitHub account. Once you create that issue, the correction will be reviewed by a staff member.

⚠️ Mobile Users: Submitting this form to create a new issue will only work with github.com, not the GitHub Mobile app.

Important: The Anthology treat PDFs as authoritative. Please use this form only to correct data that is out of line with the PDF. See our corrections guidelines if you need to change the PDF.

Title Adjust the title. Retain tags such as <fixed-case>.

Authors Adjust author names and order to match the PDF.

Abstract Correct abstract if needed. Retain XML formatting tags such as <tex-math>. You may use <b>...</b> for bold, <i>...</i> for italic, and <url>...</url> for URLs.

Verification against PDF Ensure that the new title/authors match the snapshot below. (If there is no snapshot or it is too small, consult the PDF.)

Authors concatenated from the text boxes above:

ALL author names match the snapshot above—including middle initials, hyphens, and accents.

Abstract

Linear concept vectors effectively steer LLMs, but existing methods suffer from noisy features in diverse datasets that undermine steering robustness. We propose Sparse Autoencoder-Denoised Concept Vectors (SDCV), which selectively keep the most discriminative SAE latents while reconstructing hidden representations. Our key insight is that concept-relevant signals can be explicitly separated from dataset noise by scaling up activations of top-k latents that best differentiate positive and negative samples. Applied to linear probing and difference-in-mean, SDCV consistently improves steering success rates by 4-16% across six challenging concepts, while maintaining topic relevance.

Anthology ID:: 2026.findings-eacl.40
Volume:: Findings of the Association for Computational Linguistics: EACL 2026
Month:: March
Year:: 2026
Address:: Rabat, Morocco
Editors:: Vera Demberg, Kentaro Inui, Lluís Marquez
Venue:: Findings
SIG:
Publisher:: Association for Computational Linguistics
Note:
Pages:: 797–808
Language:
URL:: https://aclanthology.org/2026.findings-eacl.40/
DOI:
Bibkey:
Cite (ACL):: Haiyan Zhao, Xuansheng Wu, Fan Yang, Bo Shen, Ninghao Liu, and Mengnan Du. 2026. Denoising Concept Vectors with Sparse Autoencoders for Improved Language Model Steering. In Findings of the Association for Computational Linguistics: EACL 2026, pages 797–808, Rabat, Morocco. Association for Computational Linguistics.
Cite (Informal):: Denoising Concept Vectors with Sparse Autoencoders for Improved Language Model Steering (Zhao et al., Findings 2026)
Copy Citation:
PDF:: https://aclanthology.org/2026.findings-eacl.40.pdf
Checklist:: 2026.findings-eacl.40.checklist.pdf

PDF Cite Search Checklist Fix data

Export citation

BibTeX
MODS XML
Endnote
Preformatted

@inproceedings{zhao-etal-2026-denoising,
    title = "Denoising Concept Vectors with Sparse Autoencoders for Improved Language Model Steering",
    author = "Zhao, Haiyan  and
      Wu, Xuansheng  and
      Yang, Fan  and
      Shen, Bo  and
      Liu, Ninghao  and
      Du, Mengnan",
    editor = "Demberg, Vera  and
      Inui, Kentaro  and
      Marquez, Llu{\'i}s",
    booktitle = "Findings of the {A}ssociation for {C}omputational {L}inguistics: {EACL} 2026",
    month = mar,
    year = "2026",
    address = "Rabat, Morocco",
    publisher = "Association for Computational Linguistics",
    url = "https://aclanthology.org/2026.findings-eacl.40/",
    pages = "797--808",
    ISBN = "979-8-89176-386-9",
    abstract = "Linear concept vectors effectively steer LLMs, but existing methods suffer from noisy features in diverse datasets that undermine steering robustness. We propose Sparse Autoencoder-Denoised Concept Vectors (SDCV), which selectively keep the most discriminative SAE latents while reconstructing hidden representations. Our key insight is that concept-relevant signals can be explicitly separated from dataset noise by scaling up activations of top-k latents that best differentiate positive and negative samples. Applied to linear probing and difference-in-mean, SDCV consistently improves steering success rates by 4-16{\%} across six challenging concepts, while maintaining topic relevance."
}

Download as File

<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="zhao-etal-2026-denoising">
    <titleInfo>
        <title>Denoising Concept Vectors with Sparse Autoencoders for Improved Language Model Steering</title>
    </titleInfo>
    <name type="personal">
        <namePart type="given">Haiyan</namePart>
        <namePart type="family">Zhao</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Xuansheng</namePart>
        <namePart type="family">Wu</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Fan</namePart>
        <namePart type="family">Yang</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Bo</namePart>
        <namePart type="family">Shen</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Ninghao</namePart>
        <namePart type="family">Liu</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Mengnan</namePart>
        <namePart type="family">Du</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <originInfo>
        <dateIssued>2026-03</dateIssued>
    </originInfo>
    <typeOfResource>text</typeOfResource>
    <relatedItem type="host">
        <titleInfo>
            <title>Findings of the Association for Computational Linguistics: EACL 2026</title>
        </titleInfo>
        <name type="personal">
            <namePart type="given">Vera</namePart>
            <namePart type="family">Demberg</namePart>
            <role>
                <roleTerm authority="marcrelator" type="text">editor</roleTerm>
            </role>
        </name>
        <name type="personal">
            <namePart type="given">Kentaro</namePart>
            <namePart type="family">Inui</namePart>
            <role>
                <roleTerm authority="marcrelator" type="text">editor</roleTerm>
            </role>
        </name>
        <name type="personal">
            <namePart type="given">Lluís</namePart>
            <namePart type="family">Marquez</namePart>
            <role>
                <roleTerm authority="marcrelator" type="text">editor</roleTerm>
            </role>
        </name>
        <originInfo>
            <publisher>Association for Computational Linguistics</publisher>
            <place>
                <placeTerm type="text">Rabat, Morocco</placeTerm>
            </place>
        </originInfo>
        <genre authority="marcgt">conference publication</genre>
        <identifier type="isbn">979-8-89176-386-9</identifier>
    </relatedItem>
    <abstract>Linear concept vectors effectively steer LLMs, but existing methods suffer from noisy features in diverse datasets that undermine steering robustness. We propose Sparse Autoencoder-Denoised Concept Vectors (SDCV), which selectively keep the most discriminative SAE latents while reconstructing hidden representations. Our key insight is that concept-relevant signals can be explicitly separated from dataset noise by scaling up activations of top-k latents that best differentiate positive and negative samples. Applied to linear probing and difference-in-mean, SDCV consistently improves steering success rates by 4-16% across six challenging concepts, while maintaining topic relevance.</abstract>
    <identifier type="citekey">zhao-etal-2026-denoising</identifier>
    <location>
        <url>https://aclanthology.org/2026.findings-eacl.40/</url>
    </location>
    <part>
        <date>2026-03</date>
        <extent unit="page">
            <start>797</start>
            <end>808</end>
        </extent>
    </part>
</mods>
</modsCollection>

Download as File

%0 Conference Proceedings
%T Denoising Concept Vectors with Sparse Autoencoders for Improved Language Model Steering
%A Zhao, Haiyan
%A Wu, Xuansheng
%A Yang, Fan
%A Shen, Bo
%A Liu, Ninghao
%A Du, Mengnan
%Y Demberg, Vera
%Y Inui, Kentaro
%Y Marquez, Lluís
%S Findings of the Association for Computational Linguistics: EACL 2026
%D 2026
%8 March
%I Association for Computational Linguistics
%C Rabat, Morocco
%@ 979-8-89176-386-9
%F zhao-etal-2026-denoising
%X Linear concept vectors effectively steer LLMs, but existing methods suffer from noisy features in diverse datasets that undermine steering robustness. We propose Sparse Autoencoder-Denoised Concept Vectors (SDCV), which selectively keep the most discriminative SAE latents while reconstructing hidden representations. Our key insight is that concept-relevant signals can be explicitly separated from dataset noise by scaling up activations of top-k latents that best differentiate positive and negative samples. Applied to linear probing and difference-in-mean, SDCV consistently improves steering success rates by 4-16% across six challenging concepts, while maintaining topic relevance.
%U https://aclanthology.org/2026.findings-eacl.40/
%P 797-808

Download as File

Markdown (Informal)

[Denoising Concept Vectors with Sparse Autoencoders for Improved Language Model Steering](https://aclanthology.org/2026.findings-eacl.40/) (Zhao et al., Findings 2026)

Denoising Concept Vectors with Sparse Autoencoders for Improved Language Model Steering (Zhao et al., Findings 2026)

ACL

Haiyan Zhao, Xuansheng Wu, Fan Yang, Bo Shen, Ninghao Liu, and Mengnan Du. 2026. Denoising Concept Vectors with Sparse Autoencoders for Improved Language Model Steering. In Findings of the Association for Computational Linguistics: EACL 2026, pages 797–808, Rabat, Morocco. Association for Computational Linguistics.