@inproceedings{mahmood-etal-2026-representation,
title = "Representation-Aware Unlearning via Activation Signatures: From Suppression to Entity-Signature Attenuation",
author = "Mahmood, Syed Naveed and
Bhuiyan, Md Rezaur Rahman and
Zaman, Tasfia and
Khondaker, Jareen Tasneem and
Sakib, Md. Sameer and
Wadith, K. M. Shadman and
Tasnim, Nazia and
Sadeque, Farig",
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.528/",
pages = "11066--11087",
ISBN = "979-8-89176-462-0",
abstract = "Entity-level unlearning is usually evaluated by what a model says: whether it stops naming the target, refuses a query, or shifts a Truth Ratio distribution. These output-level tests, however, do not show whether a subject{'}s internal representation has been attenuated. We introduce the Entity Representation Unlearning Framework (ERUF), a representation-aware framework that mines subject-specific activation signatures, suppresses the corresponding activation direction, and distills the behavior into LoRA parameters. Among evaluated baselines, ERUF is the only method that jointly achieves surface level suppression, internal attenuation, and utility preservation. On TOFU \texttt{forget10}, ERUF achieves FQ $=0.99$ and MU $=0.62$, matching reported oracle utility while approaching oracle forget quality. Across most standard foundation-model settings, ERUF maintains low leakage and low internal target activation (SMR $=0.00${--}1.10{\%}, EL10 $<0.06$, utility drift $<3\%$). On Llama-3.1-8B, adversarial entity recovery falls from 63.89{\%} to 20.15{\%}, while name-agnostic recovery decreases by 72.7{--}77.4{\%}. Joint surface/internal diagnostics further reveal scale-dependent behavior in reasoning-prior models that surface metrics alone would miss. We interpret these results as operational evidence of representation-level attenuation, not as a formal guarantee of irreversible deletion. Code is available in the repository."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="mahmood-etal-2026-representation">
<titleInfo>
<title>Representation-Aware Unlearning via Activation Signatures: From Suppression to Entity-Signature Attenuation</title>
</titleInfo>
<name type="personal">
<namePart type="given">Syed</namePart>
<namePart type="given">Naveed</namePart>
<namePart type="family">Mahmood</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Md</namePart>
<namePart type="given">Rezaur</namePart>
<namePart type="given">Rahman</namePart>
<namePart type="family">Bhuiyan</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Tasfia</namePart>
<namePart type="family">Zaman</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jareen</namePart>
<namePart type="given">Tasneem</namePart>
<namePart type="family">Khondaker</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Md.</namePart>
<namePart type="given">Sameer</namePart>
<namePart type="family">Sakib</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">K</namePart>
<namePart type="given">M</namePart>
<namePart type="given">Shadman</namePart>
<namePart type="family">Wadith</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Nazia</namePart>
<namePart type="family">Tasnim</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Farig</namePart>
<namePart type="family">Sadeque</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>Entity-level unlearning is usually evaluated by what a model says: whether it stops naming the target, refuses a query, or shifts a Truth Ratio distribution. These output-level tests, however, do not show whether a subject’s internal representation has been attenuated. We introduce the Entity Representation Unlearning Framework (ERUF), a representation-aware framework that mines subject-specific activation signatures, suppresses the corresponding activation direction, and distills the behavior into LoRA parameters. Among evaluated baselines, ERUF is the only method that jointly achieves surface level suppression, internal attenuation, and utility preservation. On TOFU forget10, ERUF achieves FQ =0.99 and MU =0.62, matching reported oracle utility while approaching oracle forget quality. Across most standard foundation-model settings, ERUF maintains low leakage and low internal target activation (SMR =0.00–1.10%, EL10 <0.06, utility drift <3%). On Llama-3.1-8B, adversarial entity recovery falls from 63.89% to 20.15%, while name-agnostic recovery decreases by 72.7–77.4%. Joint surface/internal diagnostics further reveal scale-dependent behavior in reasoning-prior models that surface metrics alone would miss. We interpret these results as operational evidence of representation-level attenuation, not as a formal guarantee of irreversible deletion. Code is available in the repository.</abstract>
<identifier type="citekey">mahmood-etal-2026-representation</identifier>
<location>
<url>https://aclanthology.org/2026.findings-emnlp.528/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>11066</start>
<end>11087</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Representation-Aware Unlearning via Activation Signatures: From Suppression to Entity-Signature Attenuation
%A Mahmood, Syed Naveed
%A Bhuiyan, Md Rezaur Rahman
%A Zaman, Tasfia
%A Khondaker, Jareen Tasneem
%A Sakib, Md. Sameer
%A Wadith, K. M. Shadman
%A Tasnim, Nazia
%A Sadeque, Farig
%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 mahmood-etal-2026-representation
%X Entity-level unlearning is usually evaluated by what a model says: whether it stops naming the target, refuses a query, or shifts a Truth Ratio distribution. These output-level tests, however, do not show whether a subject’s internal representation has been attenuated. We introduce the Entity Representation Unlearning Framework (ERUF), a representation-aware framework that mines subject-specific activation signatures, suppresses the corresponding activation direction, and distills the behavior into LoRA parameters. Among evaluated baselines, ERUF is the only method that jointly achieves surface level suppression, internal attenuation, and utility preservation. On TOFU forget10, ERUF achieves FQ =0.99 and MU =0.62, matching reported oracle utility while approaching oracle forget quality. Across most standard foundation-model settings, ERUF maintains low leakage and low internal target activation (SMR =0.00–1.10%, EL10 <0.06, utility drift <3%). On Llama-3.1-8B, adversarial entity recovery falls from 63.89% to 20.15%, while name-agnostic recovery decreases by 72.7–77.4%. Joint surface/internal diagnostics further reveal scale-dependent behavior in reasoning-prior models that surface metrics alone would miss. We interpret these results as operational evidence of representation-level attenuation, not as a formal guarantee of irreversible deletion. Code is available in the repository.
%U https://aclanthology.org/2026.findings-emnlp.528/
%P 11066-11087
Markdown (Informal)
[Representation-Aware Unlearning via Activation Signatures: From Suppression to Entity-Signature Attenuation](https://aclanthology.org/2026.findings-emnlp.528/) (Mahmood et al., Findings 2026)
ACL
- Syed Naveed Mahmood, Md Rezaur Rahman Bhuiyan, Tasfia Zaman, Jareen Tasneem Khondaker, Md. Sameer Sakib, K. M. Shadman Wadith, Nazia Tasnim, and Farig Sadeque. 2026. Representation-Aware Unlearning via Activation Signatures: From Suppression to Entity-Signature Attenuation. In Findings of the Association for Computational Linguistics: EMNLP 2026, pages 11066–11087, Budapest, Hungary. Association for Computational Linguistics.