@inproceedings{bushipaka-etal-2026-grace,
title = "{GRACE}: Gradient-guided Coreset Selection for {LLM} Unlearning",
author = "Bushipaka, Praveen and
D{'}Angelo, Andrea and
Passaro, Lucia and
Cucinotta, Tommaso",
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.1212/",
pages = "25601--25620",
ISBN = "979-8-89176-462-0",
abstract = "Machine Unlearning methods for Large Language Models typically assume pre-specified forget and retain sets. In realistic settings, however, requests may provide only a few examples of undesired behaviour, requiring forget and retain sets to be inferred from heterogeneous corpora. We study this data-selection problem and propose GRACE, a gradient-guided coreset selection method that constructs both forget and retain sets for LLM unlearning. GRACE first computes a forget direction from seed examples that elicit the undesired behaviour, then selects a compact forget coreset whose gradients approximate this direction using non-negative orthogonal matching pursuit. To preserve model utility, it selects retain examples after projecting out the forget direction and applying clustered orthogonal matching pursuit in the remaining gradient space. Across two target domains, two model families, and four unlearning algorithms, GRACE improves model utility while maintaining comparable forget quality, with particularly consistent gains over prior gradient-based selection methods."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="bushipaka-etal-2026-grace">
<titleInfo>
<title>GRACE: Gradient-guided Coreset Selection for LLM Unlearning</title>
</titleInfo>
<name type="personal">
<namePart type="given">Praveen</namePart>
<namePart type="family">Bushipaka</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Andrea</namePart>
<namePart type="family">D’Angelo</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Lucia</namePart>
<namePart type="family">Passaro</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Tommaso</namePart>
<namePart type="family">Cucinotta</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>Machine Unlearning methods for Large Language Models typically assume pre-specified forget and retain sets. In realistic settings, however, requests may provide only a few examples of undesired behaviour, requiring forget and retain sets to be inferred from heterogeneous corpora. We study this data-selection problem and propose GRACE, a gradient-guided coreset selection method that constructs both forget and retain sets for LLM unlearning. GRACE first computes a forget direction from seed examples that elicit the undesired behaviour, then selects a compact forget coreset whose gradients approximate this direction using non-negative orthogonal matching pursuit. To preserve model utility, it selects retain examples after projecting out the forget direction and applying clustered orthogonal matching pursuit in the remaining gradient space. Across two target domains, two model families, and four unlearning algorithms, GRACE improves model utility while maintaining comparable forget quality, with particularly consistent gains over prior gradient-based selection methods.</abstract>
<identifier type="citekey">bushipaka-etal-2026-grace</identifier>
<location>
<url>https://aclanthology.org/2026.findings-emnlp.1212/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>25601</start>
<end>25620</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T GRACE: Gradient-guided Coreset Selection for LLM Unlearning
%A Bushipaka, Praveen
%A D’Angelo, Andrea
%A Passaro, Lucia
%A Cucinotta, Tommaso
%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 bushipaka-etal-2026-grace
%X Machine Unlearning methods for Large Language Models typically assume pre-specified forget and retain sets. In realistic settings, however, requests may provide only a few examples of undesired behaviour, requiring forget and retain sets to be inferred from heterogeneous corpora. We study this data-selection problem and propose GRACE, a gradient-guided coreset selection method that constructs both forget and retain sets for LLM unlearning. GRACE first computes a forget direction from seed examples that elicit the undesired behaviour, then selects a compact forget coreset whose gradients approximate this direction using non-negative orthogonal matching pursuit. To preserve model utility, it selects retain examples after projecting out the forget direction and applying clustered orthogonal matching pursuit in the remaining gradient space. Across two target domains, two model families, and four unlearning algorithms, GRACE improves model utility while maintaining comparable forget quality, with particularly consistent gains over prior gradient-based selection methods.
%U https://aclanthology.org/2026.findings-emnlp.1212/
%P 25601-25620
Markdown (Informal)
[GRACE: Gradient-guided Coreset Selection for LLM Unlearning](https://aclanthology.org/2026.findings-emnlp.1212/) (Bushipaka et al., Findings 2026)
ACL
- Praveen Bushipaka, Andrea D’Angelo, Lucia Passaro, and Tommaso Cucinotta. 2026. GRACE: Gradient-guided Coreset Selection for LLM Unlearning. In Findings of the Association for Computational Linguistics: EMNLP 2026, pages 25601–25620, Budapest, Hungary. Association for Computational Linguistics.