@inproceedings{choi-etal-2026-entropy,
title = "Entropy Meets Importance: A Unified Head Importance{--}Entropy Score for Stable and Efficient Transformer Pruning",
author = "Choi, Minsik and
Son, Hyegang and
Kim, Changhoon and
Kim, Young Geun",
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.2191/",
pages = "46895--46935",
ISBN = "979-8-89176-462-0",
abstract = "Transformer-based models have achieved remarkable performance. However, their structural characteristics{---}multiple layers and attention heads{---}introduce efficiency challenges in inference and deployment. To address these challenges, various pruning methods have recently been proposed. Notably, gradient-based methods using Head Importance Scores (HIS) have gained traction for interpretability, efficiency, and ability to identify redundant heads. However, HIS alone has limitations as it captures only the gradient-driven contribution, overlooking the diversity of attention patterns. To overcome these limitations, we introduce a novel pruning criterion, HIES (Head Importance-Entropy Score), which integrates head importance scores with attention entropy, providing complementary evidence on per-head contribution. Empirically, HIES-based pruning yields a 1.13{\%} improvement in model quality and 3.52{\%} improvement in stability over the HIS-only method on average, substantially reducing the accuracy and stability cost of aggressive compression. Code will be released upon publication."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="choi-etal-2026-entropy">
<titleInfo>
<title>Entropy Meets Importance: A Unified Head Importance–Entropy Score for Stable and Efficient Transformer Pruning</title>
</titleInfo>
<name type="personal">
<namePart type="given">Minsik</namePart>
<namePart type="family">Choi</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Hyegang</namePart>
<namePart type="family">Son</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Changhoon</namePart>
<namePart type="family">Kim</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Young</namePart>
<namePart type="given">Geun</namePart>
<namePart type="family">Kim</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>Transformer-based models have achieved remarkable performance. However, their structural characteristics—multiple layers and attention heads—introduce efficiency challenges in inference and deployment. To address these challenges, various pruning methods have recently been proposed. Notably, gradient-based methods using Head Importance Scores (HIS) have gained traction for interpretability, efficiency, and ability to identify redundant heads. However, HIS alone has limitations as it captures only the gradient-driven contribution, overlooking the diversity of attention patterns. To overcome these limitations, we introduce a novel pruning criterion, HIES (Head Importance-Entropy Score), which integrates head importance scores with attention entropy, providing complementary evidence on per-head contribution. Empirically, HIES-based pruning yields a 1.13% improvement in model quality and 3.52% improvement in stability over the HIS-only method on average, substantially reducing the accuracy and stability cost of aggressive compression. Code will be released upon publication.</abstract>
<identifier type="citekey">choi-etal-2026-entropy</identifier>
<location>
<url>https://aclanthology.org/2026.findings-emnlp.2191/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>46895</start>
<end>46935</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Entropy Meets Importance: A Unified Head Importance–Entropy Score for Stable and Efficient Transformer Pruning
%A Choi, Minsik
%A Son, Hyegang
%A Kim, Changhoon
%A Kim, Young Geun
%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 choi-etal-2026-entropy
%X Transformer-based models have achieved remarkable performance. However, their structural characteristics—multiple layers and attention heads—introduce efficiency challenges in inference and deployment. To address these challenges, various pruning methods have recently been proposed. Notably, gradient-based methods using Head Importance Scores (HIS) have gained traction for interpretability, efficiency, and ability to identify redundant heads. However, HIS alone has limitations as it captures only the gradient-driven contribution, overlooking the diversity of attention patterns. To overcome these limitations, we introduce a novel pruning criterion, HIES (Head Importance-Entropy Score), which integrates head importance scores with attention entropy, providing complementary evidence on per-head contribution. Empirically, HIES-based pruning yields a 1.13% improvement in model quality and 3.52% improvement in stability over the HIS-only method on average, substantially reducing the accuracy and stability cost of aggressive compression. Code will be released upon publication.
%U https://aclanthology.org/2026.findings-emnlp.2191/
%P 46895-46935
Markdown (Informal)
[Entropy Meets Importance: A Unified Head Importance–Entropy Score for Stable and Efficient Transformer Pruning](https://aclanthology.org/2026.findings-emnlp.2191/) (Choi et al., Findings 2026)
ACL