<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="syed-etal-2024-attribution">
    <titleInfo>
        <title>Attribution Patching Outperforms Automated Circuit Discovery</title>
    </titleInfo>
    <name type="personal">
        <namePart type="given">Aaquib</namePart>
        <namePart type="family">Syed</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Can</namePart>
        <namePart type="family">Rager</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Arthur</namePart>
        <namePart type="family">Conmy</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <originInfo>
        <dateIssued>2024-11</dateIssued>
    </originInfo>
    <typeOfResource>text</typeOfResource>
    <relatedItem type="host">
        <titleInfo>
            <title>Proceedings of the 7th BlackboxNLP Workshop: Analyzing and Interpreting Neural Networks for NLP</title>
        </titleInfo>
        <name type="personal">
            <namePart type="given">Yonatan</namePart>
            <namePart type="family">Belinkov</namePart>
            <role>
                <roleTerm authority="marcrelator" type="text">editor</roleTerm>
            </role>
        </name>
        <name type="personal">
            <namePart type="given">Najoung</namePart>
            <namePart type="family">Kim</namePart>
            <role>
                <roleTerm authority="marcrelator" type="text">editor</roleTerm>
            </role>
        </name>
        <name type="personal">
            <namePart type="given">Jaap</namePart>
            <namePart type="family">Jumelet</namePart>
            <role>
                <roleTerm authority="marcrelator" type="text">editor</roleTerm>
            </role>
        </name>
        <name type="personal">
            <namePart type="given">Hosein</namePart>
            <namePart type="family">Mohebbi</namePart>
            <role>
                <roleTerm authority="marcrelator" type="text">editor</roleTerm>
            </role>
        </name>
        <name type="personal">
            <namePart type="given">Aaron</namePart>
            <namePart type="family">Mueller</namePart>
            <role>
                <roleTerm authority="marcrelator" type="text">editor</roleTerm>
            </role>
        </name>
        <name type="personal">
            <namePart type="given">Hanjie</namePart>
            <namePart type="family">Chen</namePart>
            <role>
                <roleTerm authority="marcrelator" type="text">editor</roleTerm>
            </role>
        </name>
        <originInfo>
            <publisher>Association for Computational Linguistics</publisher>
            <place>
                <placeTerm type="text">Miami, Florida, US</placeTerm>
            </place>
        </originInfo>
        <genre authority="marcgt">conference publication</genre>
    </relatedItem>
    <abstract>Automated interpretability research has recently attracted attention as a potential research direction that could scale explanations of neural network behavior to large models. Existing automated circuit discovery work applies activation patching to identify subnetworks responsible for solving specific tasks (circuits). In this work, we show that a simple method based on attribution patching outperforms all existing methods while requiring just two forward passes and a backward pass. We apply a linear approximation to activation patching to estimate the importance of each edge in the computational subgraph. Using this approximation, we prune the least important edges of the network. We survey the performance and limitations of this method, finding that averaged over all tasks our method has greater AUC from circuit recovery than other methods.</abstract>
    <identifier type="citekey">syed-etal-2024-attribution</identifier>
    <location>
        <url>https://aclanthology.org/2024.blackboxnlp-1.25</url>
    </location>
    <part>
        <date>2024-11</date>
        <extent unit="page">
            <start>407</start>
            <end>416</end>
        </extent>
    </part>
</mods>
</modsCollection>

%0 Conference Proceedings
%T Attribution Patching Outperforms Automated Circuit Discovery
%A Syed, Aaquib
%A Rager, Can
%A Conmy, Arthur
%Y Belinkov, Yonatan
%Y Kim, Najoung
%Y Jumelet, Jaap
%Y Mohebbi, Hosein
%Y Mueller, Aaron
%Y Chen, Hanjie
%S Proceedings of the 7th BlackboxNLP Workshop: Analyzing and Interpreting Neural Networks for NLP
%D 2024
%8 November
%I Association for Computational Linguistics
%C Miami, Florida, US
%F syed-etal-2024-attribution
%X Automated interpretability research has recently attracted attention as a potential research direction that could scale explanations of neural network behavior to large models. Existing automated circuit discovery work applies activation patching to identify subnetworks responsible for solving specific tasks (circuits). In this work, we show that a simple method based on attribution patching outperforms all existing methods while requiring just two forward passes and a backward pass. We apply a linear approximation to activation patching to estimate the importance of each edge in the computational subgraph. Using this approximation, we prune the least important edges of the network. We survey the performance and limitations of this method, finding that averaged over all tasks our method has greater AUC from circuit recovery than other methods.
%U https://aclanthology.org/2024.blackboxnlp-1.25
%P 407-416

Markdown (Informal)

[Attribution Patching Outperforms Automated Circuit Discovery](https://aclanthology.org/2024.blackboxnlp-1.25) (Syed et al., BlackboxNLP 2024)

• Attribution Patching Outperforms Automated Circuit Discovery (Syed et al., BlackboxNLP 2024)

ACL

• Aaquib Syed, Can Rager, and Arthur Conmy. 2024. Attribution Patching Outperforms Automated Circuit Discovery. In Proceedings of the 7th BlackboxNLP Workshop: Analyzing and Interpreting Neural Networks for NLP, pages 407–416, Miami, Florida, US. Association for Computational Linguistics.