@inproceedings{toufikuzzaman-etal-2026-blade,
title = "{BLADE}: Bilevel Low-rank Augmented-{L}agrangian Erasure for {LLM} Unlearning",
author = "Toufikuzzaman, Md and
Mousavi, Ahmad and
Lee, Dongwon",
editor = "Lee, Hung-Yi and
Dev, Sunipa and
Li, Jessy and
Elliott, Desmond",
booktitle = "Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing",
month = oct,
year = "2026",
address = "Budapest, Hungary",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.emnlp-main.953/",
pages = "21966--21987",
ISBN = "979-8-89176-459-0",
abstract = "Existing LLM unlearning methods struggle with robustness: unbounded forget losses degrade model coherence, fixed-weight balancing cannot adapt as retain difficulty shifts mid-training, and methods that work on one benchmark falter under scaling or repeated application. We propose BLADE, a constrained bilevel framework whose three mechanisms give smooth, predictable control over the optimization landscape: a clamped-entropy forget loss whose gradient is exactly zero once a token reaches sufficient uncertainty; an asymmetric augmented Lagrangian that permanently ratchets retain protection after any violation; and a bilevel structure confined to LoRA adapters that repairs retain damage before each forgetting step. BLADE dominates across three benchmark families, improving average composite scores over the strongest baselines by 6{\%} on TOFU, 9{\%} on MUSE Books, and 7{\%} on KnowUndo, and it remains stable under $4\times$ scaling and 4 sequential unlearning steps on MUSE News where the best competing method collapses entirely."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="toufikuzzaman-etal-2026-blade">
<titleInfo>
<title>BLADE: Bilevel Low-rank Augmented-Lagrangian Erasure for LLM Unlearning</title>
</titleInfo>
<name type="personal">
<namePart type="given">Md</namePart>
<namePart type="family">Toufikuzzaman</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ahmad</namePart>
<namePart type="family">Mousavi</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Dongwon</namePart>
<namePart type="family">Lee</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>Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing</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">979-8-89176-459-0</identifier>
</relatedItem>
<abstract>Existing LLM unlearning methods struggle with robustness: unbounded forget losses degrade model coherence, fixed-weight balancing cannot adapt as retain difficulty shifts mid-training, and methods that work on one benchmark falter under scaling or repeated application. We propose BLADE, a constrained bilevel framework whose three mechanisms give smooth, predictable control over the optimization landscape: a clamped-entropy forget loss whose gradient is exactly zero once a token reaches sufficient uncertainty; an asymmetric augmented Lagrangian that permanently ratchets retain protection after any violation; and a bilevel structure confined to LoRA adapters that repairs retain damage before each forgetting step. BLADE dominates across three benchmark families, improving average composite scores over the strongest baselines by 6% on TOFU, 9% on MUSE Books, and 7% on KnowUndo, and it remains stable under 4\times scaling and 4 sequential unlearning steps on MUSE News where the best competing method collapses entirely.</abstract>
<identifier type="citekey">toufikuzzaman-etal-2026-blade</identifier>
<location>
<url>https://aclanthology.org/2026.emnlp-main.953/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>21966</start>
<end>21987</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T BLADE: Bilevel Low-rank Augmented-Lagrangian Erasure for LLM Unlearning
%A Toufikuzzaman, Md
%A Mousavi, Ahmad
%A Lee, Dongwon
%Y Lee, Hung-Yi
%Y Dev, Sunipa
%Y Li, Jessy
%Y Elliott, Desmond
%S Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing
%D 2026
%8 October
%I Association for Computational Linguistics
%C Budapest, Hungary
%@ 979-8-89176-459-0
%F toufikuzzaman-etal-2026-blade
%X Existing LLM unlearning methods struggle with robustness: unbounded forget losses degrade model coherence, fixed-weight balancing cannot adapt as retain difficulty shifts mid-training, and methods that work on one benchmark falter under scaling or repeated application. We propose BLADE, a constrained bilevel framework whose three mechanisms give smooth, predictable control over the optimization landscape: a clamped-entropy forget loss whose gradient is exactly zero once a token reaches sufficient uncertainty; an asymmetric augmented Lagrangian that permanently ratchets retain protection after any violation; and a bilevel structure confined to LoRA adapters that repairs retain damage before each forgetting step. BLADE dominates across three benchmark families, improving average composite scores over the strongest baselines by 6% on TOFU, 9% on MUSE Books, and 7% on KnowUndo, and it remains stable under 4\times scaling and 4 sequential unlearning steps on MUSE News where the best competing method collapses entirely.
%U https://aclanthology.org/2026.emnlp-main.953/
%P 21966-21987
Markdown (Informal)
[BLADE: Bilevel Low-rank Augmented-Lagrangian Erasure for LLM Unlearning](https://aclanthology.org/2026.emnlp-main.953/) (Toufikuzzaman et al., EMNLP 2026)
ACL