@inproceedings{ju-etal-2022-logic,
title = "Logic Traps in Evaluating Attribution Scores",
author = "Ju, Yiming and
Zhang, Yuanzhe and
Yang, Zhao and
Jiang, Zhongtao and
Liu, Kang and
Zhao, Jun",
editor = "Muresan, Smaranda and
Nakov, Preslav and
Villavicencio, Aline",
booktitle = "Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)",
month = may,
year = "2022",
address = "Dublin, Ireland",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2022.acl-long.407",
doi = "10.18653/v1/2022.acl-long.407",
pages = "5911--5922",
abstract = "Modern deep learning models are notoriously opaque, which has motivated the development of methods for interpreting how deep models predict. This goal is usually approached with attribution method, which assesses the influence of features on model predictions. As an explanation method, the evaluation criteria of attribution methods is how accurately it reflects the actual reasoning process of the model (faithfulness). Meanwhile, since the reasoning process of deep models is inaccessible, researchers design various evaluation methods to demonstrate their arguments. However, some crucial logic traps in these evaluation methods are ignored in most works, causing inaccurate evaluation and unfair comparison. This paper systematically reviews existing methods for evaluating attribution scores and summarizes the logic traps in these methods. We further conduct experiments to demonstrate the existence of each logic trap. Through both theoretical and experimental analysis, we hope to increase attention on the inaccurate evaluation of attribution scores. Moreover, with this paper, we suggest stopping focusing on improving performance under unreliable evaluation systems and starting efforts on reducing the impact of proposed logic traps.",
}
<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="ju-etal-2022-logic">
<titleInfo>
<title>Logic Traps in Evaluating Attribution Scores</title>
</titleInfo>
<name type="personal">
<namePart type="given">Yiming</namePart>
<namePart type="family">Ju</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yuanzhe</namePart>
<namePart type="family">Zhang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Zhao</namePart>
<namePart type="family">Yang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Zhongtao</namePart>
<namePart type="family">Jiang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Kang</namePart>
<namePart type="family">Liu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jun</namePart>
<namePart type="family">Zhao</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2022-05</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)</title>
</titleInfo>
<name type="personal">
<namePart type="given">Smaranda</namePart>
<namePart type="family">Muresan</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Preslav</namePart>
<namePart type="family">Nakov</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Aline</namePart>
<namePart type="family">Villavicencio</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">Dublin, Ireland</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
</relatedItem>
<abstract>Modern deep learning models are notoriously opaque, which has motivated the development of methods for interpreting how deep models predict. This goal is usually approached with attribution method, which assesses the influence of features on model predictions. As an explanation method, the evaluation criteria of attribution methods is how accurately it reflects the actual reasoning process of the model (faithfulness). Meanwhile, since the reasoning process of deep models is inaccessible, researchers design various evaluation methods to demonstrate their arguments. However, some crucial logic traps in these evaluation methods are ignored in most works, causing inaccurate evaluation and unfair comparison. This paper systematically reviews existing methods for evaluating attribution scores and summarizes the logic traps in these methods. We further conduct experiments to demonstrate the existence of each logic trap. Through both theoretical and experimental analysis, we hope to increase attention on the inaccurate evaluation of attribution scores. Moreover, with this paper, we suggest stopping focusing on improving performance under unreliable evaluation systems and starting efforts on reducing the impact of proposed logic traps.</abstract>
<identifier type="citekey">ju-etal-2022-logic</identifier>
<identifier type="doi">10.18653/v1/2022.acl-long.407</identifier>
<location>
<url>https://aclanthology.org/2022.acl-long.407</url>
</location>
<part>
<date>2022-05</date>
<extent unit="page">
<start>5911</start>
<end>5922</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Logic Traps in Evaluating Attribution Scores
%A Ju, Yiming
%A Zhang, Yuanzhe
%A Yang, Zhao
%A Jiang, Zhongtao
%A Liu, Kang
%A Zhao, Jun
%Y Muresan, Smaranda
%Y Nakov, Preslav
%Y Villavicencio, Aline
%S Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)
%D 2022
%8 May
%I Association for Computational Linguistics
%C Dublin, Ireland
%F ju-etal-2022-logic
%X Modern deep learning models are notoriously opaque, which has motivated the development of methods for interpreting how deep models predict. This goal is usually approached with attribution method, which assesses the influence of features on model predictions. As an explanation method, the evaluation criteria of attribution methods is how accurately it reflects the actual reasoning process of the model (faithfulness). Meanwhile, since the reasoning process of deep models is inaccessible, researchers design various evaluation methods to demonstrate their arguments. However, some crucial logic traps in these evaluation methods are ignored in most works, causing inaccurate evaluation and unfair comparison. This paper systematically reviews existing methods for evaluating attribution scores and summarizes the logic traps in these methods. We further conduct experiments to demonstrate the existence of each logic trap. Through both theoretical and experimental analysis, we hope to increase attention on the inaccurate evaluation of attribution scores. Moreover, with this paper, we suggest stopping focusing on improving performance under unreliable evaluation systems and starting efforts on reducing the impact of proposed logic traps.
%R 10.18653/v1/2022.acl-long.407
%U https://aclanthology.org/2022.acl-long.407
%U https://doi.org/10.18653/v1/2022.acl-long.407
%P 5911-5922
Markdown (Informal)
[Logic Traps in Evaluating Attribution Scores](https://aclanthology.org/2022.acl-long.407) (Ju et al., ACL 2022)
ACL
- Yiming Ju, Yuanzhe Zhang, Zhao Yang, Zhongtao Jiang, Kang Liu, and Jun Zhao. 2022. Logic Traps in Evaluating Attribution Scores. In Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 5911–5922, Dublin, Ireland. Association for Computational Linguistics.