@inproceedings{zylberajch-etal-2021-hildif,
title = "{HILDIF}: {I}nteractive Debugging of {NLI} Models Using Influence Functions",
author = "Zylberajch, Hugo and
Lertvittayakumjorn, Piyawat and
Toni, Francesca",
editor = {Brantley, Kiant{\'e} and
Dan, Soham and
Gurevych, Iryna and
Lee, Ji-Ung and
Radlinski, Filip and
Sch{\"u}tze, Hinrich and
Simpson, Edwin and
Yu, Lili},
booktitle = "Proceedings of the First Workshop on Interactive Learning for Natural Language Processing",
month = aug,
year = "2021",
address = "Online",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2021.internlp-1.1",
doi = "10.18653/v1/2021.internlp-1.1",
pages = "1--6",
abstract = "Biases and artifacts in training data can cause unwelcome behavior in text classifiers (such as shallow pattern matching), leading to lack of generalizability. One solution to this problem is to include users in the loop and leverage their feedback to improve models. We propose a novel explanatory debugging pipeline called HILDIF, enabling humans to improve deep text classifiers using influence functions as an explanation method. We experiment on the Natural Language Inference (NLI) task, showing that HILDIF can effectively alleviate artifact problems in fine-tuned BERT models and result in increased model generalizability.",
}
<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="zylberajch-etal-2021-hildif">
<titleInfo>
<title>HILDIF: Interactive Debugging of NLI Models Using Influence Functions</title>
</titleInfo>
<name type="personal">
<namePart type="given">Hugo</namePart>
<namePart type="family">Zylberajch</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Piyawat</namePart>
<namePart type="family">Lertvittayakumjorn</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Francesca</namePart>
<namePart type="family">Toni</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2021-08</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Proceedings of the First Workshop on Interactive Learning for Natural Language Processing</title>
</titleInfo>
<name type="personal">
<namePart type="given">Kianté</namePart>
<namePart type="family">Brantley</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Soham</namePart>
<namePart type="family">Dan</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Iryna</namePart>
<namePart type="family">Gurevych</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ji-Ung</namePart>
<namePart type="family">Lee</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Filip</namePart>
<namePart type="family">Radlinski</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Hinrich</namePart>
<namePart type="family">Schütze</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Edwin</namePart>
<namePart type="family">Simpson</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Lili</namePart>
<namePart type="family">Yu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">Online</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
</relatedItem>
<abstract>Biases and artifacts in training data can cause unwelcome behavior in text classifiers (such as shallow pattern matching), leading to lack of generalizability. One solution to this problem is to include users in the loop and leverage their feedback to improve models. We propose a novel explanatory debugging pipeline called HILDIF, enabling humans to improve deep text classifiers using influence functions as an explanation method. We experiment on the Natural Language Inference (NLI) task, showing that HILDIF can effectively alleviate artifact problems in fine-tuned BERT models and result in increased model generalizability.</abstract>
<identifier type="citekey">zylberajch-etal-2021-hildif</identifier>
<identifier type="doi">10.18653/v1/2021.internlp-1.1</identifier>
<location>
<url>https://aclanthology.org/2021.internlp-1.1</url>
</location>
<part>
<date>2021-08</date>
<extent unit="page">
<start>1</start>
<end>6</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T HILDIF: Interactive Debugging of NLI Models Using Influence Functions
%A Zylberajch, Hugo
%A Lertvittayakumjorn, Piyawat
%A Toni, Francesca
%Y Brantley, Kianté
%Y Dan, Soham
%Y Gurevych, Iryna
%Y Lee, Ji-Ung
%Y Radlinski, Filip
%Y Schütze, Hinrich
%Y Simpson, Edwin
%Y Yu, Lili
%S Proceedings of the First Workshop on Interactive Learning for Natural Language Processing
%D 2021
%8 August
%I Association for Computational Linguistics
%C Online
%F zylberajch-etal-2021-hildif
%X Biases and artifacts in training data can cause unwelcome behavior in text classifiers (such as shallow pattern matching), leading to lack of generalizability. One solution to this problem is to include users in the loop and leverage their feedback to improve models. We propose a novel explanatory debugging pipeline called HILDIF, enabling humans to improve deep text classifiers using influence functions as an explanation method. We experiment on the Natural Language Inference (NLI) task, showing that HILDIF can effectively alleviate artifact problems in fine-tuned BERT models and result in increased model generalizability.
%R 10.18653/v1/2021.internlp-1.1
%U https://aclanthology.org/2021.internlp-1.1
%U https://doi.org/10.18653/v1/2021.internlp-1.1
%P 1-6
Markdown (Informal)
[HILDIF: Interactive Debugging of NLI Models Using Influence Functions](https://aclanthology.org/2021.internlp-1.1) (Zylberajch et al., InterNLP 2021)
ACL