DMDIntel: Interpreting Large Language Models via Dynamic Mode Decomposition

Amogh Joshi; Animesh Mukherjee; Sergey Utyuzhnikov

DMDIntel: Interpreting Large Language Models via Dynamic Mode Decomposition

Amogh Joshi, Animesh Mukherjee, Sergey Utyuzhnikov

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 ... for bold, ... for italic, ... for underline, <sc>...</sc> for small-caps, <tt>...<tt> for typewriter text, <url>...</url> for URLs, <a href=...> for hyperlinks, and <par/> for paragraph breaks.

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

In this work, we introduce DMDIntel which uses *dynamic mode decomposition* (DMD) to make the predictions made by LLMs in a classification task interpretable. It develops an input attribution pipeline, that first decomposes the hidden states of an LLM into prominent patterns, also known as *modes*, and then associates ranks to the input tokens based on the projection values on those modes. Rigorous experiments across three datasets and three model families consistently show that the ranked attribution of input tokens obtained using DMDIntel by far outperforms state-of-the-art techniques such as principal component analysis, integrated gradients and SHAP.

Anthology ID:: 2026.findings-emnlp.1545
Volume:: Findings of the Association for Computational Linguistics: EMNLP 2026
Month:: October
Year:: 2026
Address:: Budapest, Hungary
Editors:: Hung-Yi Lee, Sunipa Dev, Jessy Li, Desmond Elliott
Venue:: Findings
SIG:: SIGDAT
Publisher:: Association for Computational Linguistics
Note:
Pages:: 32870–32889
Language:
URL:: https://aclanthology.org/2026.findings-emnlp.1545/
DOI:
Bibkey:
Cite (ACL):: Amogh Joshi, Animesh Mukherjee, and Sergey Utyuzhnikov. 2026. DMDIntel: Interpreting Large Language Models via Dynamic Mode Decomposition. In Findings of the Association for Computational Linguistics: EMNLP 2026, pages 32870–32889, Budapest, Hungary. Association for Computational Linguistics.
Cite (Informal):: DMDIntel: Interpreting Large Language Models via Dynamic Mode Decomposition (Joshi et al., Findings 2026)
Copy Citation:
PDF:: https://aclanthology.org/2026.findings-emnlp.1545.pdf

PDF Cite Search Fix data

Export citation

BibTeX
MODS XML
Endnote
Preformatted

@inproceedings{joshi-etal-2026-dmdintel,
    title = "{DMDI}ntel: Interpreting Large Language Models via Dynamic Mode Decomposition",
    author = "Joshi, Amogh  and
      Mukherjee, Animesh  and
      Utyuzhnikov, Sergey",
    editor = "Lee, Hung-Yi  and
      Dev, Sunipa  and
      Li, Jessy  and
      Elliott, Desmond",
    booktitle = "Findings of the {A}ssociation for {C}omputational {L}inguistics: {EMNLP} 2026",
    month = oct,
    year = "2026",
    address = "Budapest, Hungary",
    publisher = "Association for Computational Linguistics",
    url = "https://aclanthology.org/2026.findings-emnlp.1545/",
    pages = "32870--32889",
    ISBN = "979-8-89176-462-0",
    abstract = "In this work, we introduce DMDIntel which uses *dynamic mode decomposition* (DMD) to make the predictions made by LLMs in a classification task interpretable. It develops an input attribution pipeline, that first decomposes the hidden states of an LLM into prominent patterns, also known as *modes*, and then associates ranks to the input tokens based on the projection values on those modes. Rigorous experiments across three datasets and three model families consistently show that the ranked attribution of input tokens obtained using DMDIntel by far outperforms state-of-the-art techniques such as principal component analysis, integrated gradients and SHAP."
}

Download as File

<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="joshi-etal-2026-dmdintel">
    <titleInfo>
        <title>DMDIntel: Interpreting Large Language Models via Dynamic Mode Decomposition</title>
    </titleInfo>
    <name type="personal">
        <namePart type="given">Amogh</namePart>
        <namePart type="family">Joshi</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Animesh</namePart>
        <namePart type="family">Mukherjee</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Sergey</namePart>
        <namePart type="family">Utyuzhnikov</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>Findings of the Association for Computational Linguistics: EMNLP 2026</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">979-8-89176-462-0</identifier>
    </relatedItem>
    <abstract>In this work, we introduce DMDIntel which uses *dynamic mode decomposition* (DMD) to make the predictions made by LLMs in a classification task interpretable. It develops an input attribution pipeline, that first decomposes the hidden states of an LLM into prominent patterns, also known as *modes*, and then associates ranks to the input tokens based on the projection values on those modes. Rigorous experiments across three datasets and three model families consistently show that the ranked attribution of input tokens obtained using DMDIntel by far outperforms state-of-the-art techniques such as principal component analysis, integrated gradients and SHAP.</abstract>
    <identifier type="citekey">joshi-etal-2026-dmdintel</identifier>
    <location>
        <url>https://aclanthology.org/2026.findings-emnlp.1545/</url>
    </location>
    <part>
        <date>2026-10</date>
        <extent unit="page">
            <start>32870</start>
            <end>32889</end>
        </extent>
    </part>
</mods>
</modsCollection>

Download as File

%0 Conference Proceedings
%T DMDIntel: Interpreting Large Language Models via Dynamic Mode Decomposition
%A Joshi, Amogh
%A Mukherjee, Animesh
%A Utyuzhnikov, Sergey
%Y Lee, Hung-Yi
%Y Dev, Sunipa
%Y Li, Jessy
%Y Elliott, Desmond
%S Findings of the Association for Computational Linguistics: EMNLP 2026
%D 2026
%8 October
%I Association for Computational Linguistics
%C Budapest, Hungary
%@ 979-8-89176-462-0
%F joshi-etal-2026-dmdintel
%X In this work, we introduce DMDIntel which uses *dynamic mode decomposition* (DMD) to make the predictions made by LLMs in a classification task interpretable. It develops an input attribution pipeline, that first decomposes the hidden states of an LLM into prominent patterns, also known as *modes*, and then associates ranks to the input tokens based on the projection values on those modes. Rigorous experiments across three datasets and three model families consistently show that the ranked attribution of input tokens obtained using DMDIntel by far outperforms state-of-the-art techniques such as principal component analysis, integrated gradients and SHAP.
%U https://aclanthology.org/2026.findings-emnlp.1545/
%P 32870-32889

Download as File

Markdown (Informal)

[DMDIntel: Interpreting Large Language Models via Dynamic Mode Decomposition](https://aclanthology.org/2026.findings-emnlp.1545/) (Joshi et al., Findings 2026)

DMDIntel: Interpreting Large Language Models via Dynamic Mode Decomposition (Joshi et al., Findings 2026)

ACL

Amogh Joshi, Animesh Mukherjee, and Sergey Utyuzhnikov. 2026. DMDIntel: Interpreting Large Language Models via Dynamic Mode Decomposition. In Findings of the Association for Computational Linguistics: EMNLP 2026, pages 32870–32889, Budapest, Hungary. Association for Computational Linguistics.