@inproceedings{li-etal-2026-sgpvt,
title = "{SGPVT}: Self-Generated Proximal Visual Tokens for Mitigating Proximal Collateral Damage in {MLLM} Unlearning",
author = "Li, Jiaqi and
Zhang, Zhijing and
Geng, Jiahui and
Bi, Sheng and
Zhang, Chuanyi and
Liu, Fan and
Qi, Guilin",
editor = "Liakata, Maria and
Moreira, Viviane P. and
Zhang, Jiajun and
Jurgens, David",
booktitle = "Proceedings of the 64th Annual Meeting of the {A}ssociation for {C}omputational {L}inguistics (Volume 1: Long Papers)",
month = jul,
year = "2026",
address = "San Diego, California, United States",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.acl-long.442/",
doi = "10.18653/v1/2026.acl-long.442",
pages = "9744--9763",
ISBN = "979-8-89176-390-6",
abstract = "Machine unlearning in multimodal large language models (MLLMs) aims to remove specific concepts while preserving overall utility. However, existing approaches focus primarily on general utility metrics, overlooking the preservation of semantically related concepts. We present the first systematic analysis of this proximal collateral damage, revealing that forgetting vulnerability correlates strongly with visual embedding similarity in a smooth gradient across the semantic space. Based on this insight, we propose a novel unlearning framework that introduces Self-Generated Proximal Visual Tokens (SGPVTs), which are synthetically perturbed visual representations around the target concept. Our method employs an adaptive cosine-band curriculum with a dual-stream objective: forgetting the target via gradient ascent while distilling knowledge from a frozen teacher model into proximal tokens to prevent degradation. Extensive experiments demonstrate that our approach significantly outperforms existing methods in preserving semantically related concepts while achieving effective target unlearning, eliminating the need for manual retention set curation. Our source code will be released in the near future."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="li-etal-2026-sgpvt">
<titleInfo>
<title>SGPVT: Self-Generated Proximal Visual Tokens for Mitigating Proximal Collateral Damage in MLLM Unlearning</title>
</titleInfo>
<name type="personal">
<namePart type="given">Jiaqi</namePart>
<namePart type="family">Li</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Zhijing</namePart>
<namePart type="family">Zhang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jiahui</namePart>
<namePart type="family">Geng</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Sheng</namePart>
<namePart type="family">Bi</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Chuanyi</namePart>
<namePart type="family">Zhang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Fan</namePart>
<namePart type="family">Liu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Guilin</namePart>
<namePart type="family">Qi</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2026-07</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)</title>
</titleInfo>
<name type="personal">
<namePart type="given">Maria</namePart>
<namePart type="family">Liakata</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Viviane</namePart>
<namePart type="given">P</namePart>
<namePart type="family">Moreira</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jiajun</namePart>
<namePart type="family">Zhang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">David</namePart>
<namePart type="family">Jurgens</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">San Diego, California, United States</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
<identifier type="isbn">979-8-89176-390-6</identifier>
</relatedItem>
<abstract>Machine unlearning in multimodal large language models (MLLMs) aims to remove specific concepts while preserving overall utility. However, existing approaches focus primarily on general utility metrics, overlooking the preservation of semantically related concepts. We present the first systematic analysis of this proximal collateral damage, revealing that forgetting vulnerability correlates strongly with visual embedding similarity in a smooth gradient across the semantic space. Based on this insight, we propose a novel unlearning framework that introduces Self-Generated Proximal Visual Tokens (SGPVTs), which are synthetically perturbed visual representations around the target concept. Our method employs an adaptive cosine-band curriculum with a dual-stream objective: forgetting the target via gradient ascent while distilling knowledge from a frozen teacher model into proximal tokens to prevent degradation. Extensive experiments demonstrate that our approach significantly outperforms existing methods in preserving semantically related concepts while achieving effective target unlearning, eliminating the need for manual retention set curation. Our source code will be released in the near future.</abstract>
<identifier type="citekey">li-etal-2026-sgpvt</identifier>
<identifier type="doi">10.18653/v1/2026.acl-long.442</identifier>
<location>
<url>https://aclanthology.org/2026.acl-long.442/</url>
</location>
<part>
<date>2026-07</date>
<extent unit="page">
<start>9744</start>
<end>9763</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T SGPVT: Self-Generated Proximal Visual Tokens for Mitigating Proximal Collateral Damage in MLLM Unlearning
%A Li, Jiaqi
%A Zhang, Zhijing
%A Geng, Jiahui
%A Bi, Sheng
%A Zhang, Chuanyi
%A Liu, Fan
%A Qi, Guilin
%Y Liakata, Maria
%Y Moreira, Viviane P.
%Y Zhang, Jiajun
%Y Jurgens, David
%S Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)
%D 2026
%8 July
%I Association for Computational Linguistics
%C San Diego, California, United States
%@ 979-8-89176-390-6
%F li-etal-2026-sgpvt
%X Machine unlearning in multimodal large language models (MLLMs) aims to remove specific concepts while preserving overall utility. However, existing approaches focus primarily on general utility metrics, overlooking the preservation of semantically related concepts. We present the first systematic analysis of this proximal collateral damage, revealing that forgetting vulnerability correlates strongly with visual embedding similarity in a smooth gradient across the semantic space. Based on this insight, we propose a novel unlearning framework that introduces Self-Generated Proximal Visual Tokens (SGPVTs), which are synthetically perturbed visual representations around the target concept. Our method employs an adaptive cosine-band curriculum with a dual-stream objective: forgetting the target via gradient ascent while distilling knowledge from a frozen teacher model into proximal tokens to prevent degradation. Extensive experiments demonstrate that our approach significantly outperforms existing methods in preserving semantically related concepts while achieving effective target unlearning, eliminating the need for manual retention set curation. Our source code will be released in the near future.
%R 10.18653/v1/2026.acl-long.442
%U https://aclanthology.org/2026.acl-long.442/
%U https://doi.org/10.18653/v1/2026.acl-long.442
%P 9744-9763
Markdown (Informal)
[SGPVT: Self-Generated Proximal Visual Tokens for Mitigating Proximal Collateral Damage in MLLM Unlearning](https://aclanthology.org/2026.acl-long.442/) (Li et al., ACL 2026)
ACL
- Jiaqi Li, Zhijing Zhang, Jiahui Geng, Sheng Bi, Chuanyi Zhang, Fan Liu, and Guilin Qi. 2026. SGPVT: Self-Generated Proximal Visual Tokens for Mitigating Proximal Collateral Damage in MLLM Unlearning. In Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 9744–9763, San Diego, California, United States. Association for Computational Linguistics.