Cross-Modal Attention Guided Unlearning in Vision-Language Models
Listen
Radio episode about this paper
Transcript
Introduction to the show: ident: AI Radio. Generated commentary on the latest Artificial Intelligence papers.
Tom: I'm Tom.
Jane: And I'm Jane, here with Lu, Meng and our in-house large language model, Lalam.
Tom: We're looking at the paper: "Cross-Modal Attention Guided Unlearning in Vision-Language Models".
Paper summary: Tom: So we're looking at this paper called "Cross-Modal Attention Guided Unlearning in Vision-Language Models." It looks like they're tackling the problem of forgetting private data from these big vision and language models, specifically for visual question answering tasks.
Jane: Right, so the core idea here seems to be that instead of retraining the whole massive model, they want a way to selectively forget certain information linked to specific queries by looking at how the visual and text parts interact during processing.
Lu: Exactly. They're proposing this framework called CAGUL, which uses cross-modal attention scores between the visual and text inputs to guide where they encode the unlearning information in the visual tokens that are least important for a particular query.
Meng: Wait, so how does it actually work? I mean, if we don't retrain the main model parameters, what are we actually modifying? Does this just change the input embeddings before they go to the language model?
Tom: It’s not just changing the input. They have these external modules that are trained separately. There's a discriminator module that checks if an image is paired with a private query, and then there's this visual token encoder that transforms the visual tokens based on those attention scores.
Jane: So, to put it plainly, they use the attention map between the picture and the question to decide which parts of the picture are least relevant when you want to forget something. They conjecture that information in those least attended tokens is useful for unlearning private or non-private queries.
Lu: That's what they're doing. They’re using this idea from how VLMs extract details, where they compress visual info into a small set of highly attended tokens, and applying the opposite logic—focusing on the least attended ones for forgetting.
Meng: Okay, so it’s a selection process based on attention scores leading to an encoding step. But what about training? How do we train these external modules without messing up the billions of parameters in the VLM itself?
Tom: They train two main things: the discriminator module and this visual token encoder. The discriminator is trained with a classification objective, and the encoder is trained using specific loss objectives for both forgetting and retaining data separately.
Jane: For forgetting, they use something called Preference Optimization to align the language model's outputs for forget queries with refusal responses instead of the actual correct answer. That’s a clever way to signal what we want to erase.
Lu: And for retaining data, they have an objective that tries to maintain the base model’s predictive utility, which is important because they don't just want to erase everything; they want to keep general knowledge intact.
Meng: I see the trade-off there. So you're trying to find this sweet spot where you forget the private data without destroying the model’s ability to answer normal questions correctly? That sounds like it could be a big deal for production systems.
Paper summary: Tom: Experimental results show that CAGUL performs better than finetuning baselines on forgetting quality and even keeps the retain utility of the finetuned base model. And what's cool is they don't alter any of the pre-trained VLM parameters or incur retraining costs.
Jane: That’s significant because it means you can exercise that right to be forgotten for specific subsets of data without having to spend a lot of time and compute on full finetuning. And they kept the computational efficiency quite high, training far fewer parameters than other methods.
Lu: The ablation study shows that while CAGUL can get good trade-offs between forgetting and retaining metrics without the discriminator, its generalization ability suffers if you remove it. But without this encoder module, forget and retain performance drops significantly.
Meng: So it seems the discriminator is crucial for knowing *when* to apply the token selection mechanism, and the encoder is necessary to actually embed that unlearning signal into those selected tokens. That’s a lot of moving parts to get right at scale.
Tom: This whole CAGUL framework is essentially a way for individual users to exercise the right to be forgotten on subsets of their sensitive queries within these vision-language models. It’s a novel setting for machine unlearning in this domain.
Jane: So, the paper "Cross-Modal Attention Guided Unlearning in Vision-Language Models" proposes using cross-modal attention to guide the encoding of unlearning information into low-importance visual tokens for relevant queries.
Lu: That's the main mechanism they are investigating, focusing on how those attention scores between vision and text dictate which visual tokens get modified for forgetting.
Meng: It’s interesting because it leverages existing attention mechanisms instead of completely rebuilding the model training loop or fine-tuning every layer we have to touch. That makes deployment much more practical.
Tom: The authors show this method works on the FIUBench dataset, using two different open-source vision-language models, and they show it achieves good trade-offs compared to finetuning methods without altering the pre-trained model parameters or retraining costs.
Jane: So what this means for us is that we can potentially make these powerful multimodal models more private and compliant with privacy regulations by using targeted unlearning instead of broad fine-tuning.
Lu: It shifts the focus from massive parameter updates to targeted modifications guided by attention patterns, which is a different direction in how we think about model modification.
Meng: If this scales well on real traffic, it could be a huge win for deploying these models where data privacy is a major concern, because the cost of unlearning seems much lower.
Tom: The conclusion of this work is that CAGUL provides an effective way to handle machine unlearning in vision-language models by using a classifier to identify images from the forget set, an attention-based selection mechanism, and an encoder for embedding information into low-attention visual tokens.
Jane: Essentially, they’ve found a novel setting where users can forget specific subsets of attributes available to the model by leveraging cross-modal attention. It’s a practical approach to privacy in multimodal AI.
Conclusion: Tom: So, we're wrapping up on "Cross-Modal Attention Guided Unlearning in Vision-Language Models." Basically, they showed how you can selectively forget data from these big vision and language models without having to retrain the whole thing from scratch.
Jane: Right, so the title is Cross-Modal Attention Guided Unlearning in Vision-Language Models. It’s about using that attention score between the picture and the text to guide where they put information when you want to forget something.
Lu: They use a discriminator module and a visual token encoder to do this selective forgetting based on how important those tokens are for the specific query you’re asking about.
Meng: I just want to know what this means in practice, Tom. If someone wants to forget their private data, does this mean we can just point at the picture and type the query? Or is there a more complex setup?
Tom: It's more complex than that. You need that discriminator module to first figure out if the image is actually paired with a private query before you even look at those attention scores for token selection.
Jane: And they showed it works on two different vision and language models, meaning this approach isn't tied to just one specific model architecture.
Lu: The key result is that this method achieves a good balance between forgetting the data and keeping the model smart enough to answer regular questions, which is what we call retain utility.
Meng: That balance sounds important because usually when you try to erase something, you either lose too much general knowledge or you forget too much of what's actually private.
Tom: Exactly. They found a sweet spot where it performs better than traditional finetuning methods on the forgetting side while preserving that retained utility.
Jane: So the implication is that for anyone dealing with sensitive data, this offers a new way to exercise the right to be forgotten on specific parts of what a multimodal AI knows.
Lu: It’s about moving away from massive parameter updates toward targeted modifications guided by how the model processes information cross-modally.
Meng: That makes sense for deployment because it doesn't require retraining the entire pre-trained model, which saves a ton of computational power and time at scale.
Tom: It’s definitely more efficient than those full finetuning baselines they compared it against, especially when you look at the training costs.
Jane: And this sets up a lot of interesting work for future research on how to make these models more private without sacrificing their overall usefulness for general tasks.
Karuna Bhaila, Aneesh Komanduri, Minh-Hao Van, Xintao Wu
University of Arkansas
cs.CV
Submitted: 2025-10-08
Updated: 2026-11-14
License: http://creativecommons.org/licenses/by/4.0/
Importance score: 83/100
The gist: Vision-Language Models (VLMs) have demonstrated immense capabilities in multi-modal understanding and inference tasks such as Visual Question Answering (VQA)…
Key concepts
- Cross-Modal Attention Scores
- These are the numerical scores calculated between visual input tokens (images) and text input tokens (queries). They measure how closely related a specific part of an image is to a specific word in the prompt. The method uses these scores to decide which visual information is most important for unlearning.
- Discriminator Module
- This module acts as a classifier that determines whether a given visual input is associated with a query about private data or not. It helps the framework identify which images need to be targeted for the unlearning process, guiding the subsequent token selection.
- Visual Token Encoder (Fψ)
- This component takes visual tokens deemed important for unlearning and linearly transforms them. By focusing on tokens with low cross-modal attention scores, this encoder embeds the specific information needed to forget data without significantly disrupting the model's overall behavior.
Terminology
Summary
Vision-Language Models (VLMs) have demonstrated immense capabilities in multi-modal understanding and inference tasks such as Visual Question Answering (VQA), which requires models to infer outputs based on visual and textual context simultaneously. p. 0 Cross-Modal Attention Guided Unlearning in Vision-Language Models
Karuna Bhaila, Aneesh Komanduri, Minh-Hao Van, Xintao Wu
Department of Electrical Engineering and Computer Science
University of Arkansas
{kbhaila,akomandu,haovan,xintaowu}@uark. edu
Abstract
Vision-Language Models (VLMs) have demonstrated immense capabilities in multi-modal understanding and inference tasks such as Visual Question Answering (VQA), which
requires models to infer outputs based on visual and textual
context simultaneously. Based on this problem definition, we propose a
lightweight and efficient method to achieve VLM unlearning. Our method is motivated by the relationship between
visual and text tokens represented as cross-modal attention
scores computed between the two modes of input. We explore the role
of visual tokens for output generation in VLMs using crossmodal attention and utilize it to formulate Cross-Modal Attention Guided Unlearning (CAGUL), a lightweight and efficient VLM unlearning framework. In contrast to computationally expensive model finetuning methods, CAGUL utilizes external modules to encode unlearning information in
visual tokens of low importance for relevant queries. In contrast to computationally expensive model finetuning methods, CAGUL utilizes external modules to encode unlearning information in
visual tokens of low importance for relevant queries. We find
that the transformed visual tokens not only prevent leakage
but also retain reference model behavior.Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Framework Components
The CAGUL framework is composed of two main components: a discriminator module to detect whether or not a given visual input is paired with a query about private data, and a cross-modal attention-based visual token selection mechanism. Then, for visual inputs predicted to be paired with private data, we use an MLP encoder to linearly transform the visual tokens with the lowest cross-modal attention scores. p. 1 First, a discriminator module detects whether or not a
given visual input is paired with a query about private data. Then, for visual inputs predicted to be paired with private
data, we use an MLP encoder to linearly transform the visual
tokens with the lowest cross-modal attention scores. We then replace the k image token embeddings in Zv
with the transformed representation as follows:
Zˆ
v,i =
Fψ(Zv,i) if i ∈ K
Zv,i otherwise (8)
where Zˆ
v is the set of final visual token embeddings. The
final input to the LLM can be represented as follows:
Yˆ = LLMθ(Zˆ
v, Zq) (9)
Unlearning Objective.Where the paper says it
Where the paper says it
Unlearning Objectives
The framework has two trainable modules: the discriminator Cϕ and the visual token encoder Fψ. The discriminator is trained with a classification objective defined as Lbce = EZv log pCϕ(lZv). p. 3 Our CAGUL framework has two
trainable modules, the discriminator Cϕ and the visual token
encoder Fψ. The discriminator is trained with the following
classification objective
Lbce = EZv
log pCϕ
(l|Zv)
(10)
The encoder is trained with specific unlearning objectives
defined on the forget and retain sets separately. Our CAGUL framework has two
trainable modules, the discriminator Cϕ and the visual token
encoder Fψ. The discriminator is trained with the following
classification objective
Lbce = EZv
log pCϕ
(l|Zv)
(10)
The encoder is trained with specific unlearning objectives
defined on the forget and retain sets separately. Our CAGUL framework has two
trainable modules, the discriminator Cϕ and the visual token
encoder Fψ. The discriminator is trained with the following
classification objective
Lbce = EZv
log pCϕ
(l|Zv)
(10)
The encoder is trained with specific unlearning objectives
defined on the forget and retain sets separately. The discriminator is trained with the following
classification objective
Lbce = EZv
log pCϕ
(l|Zv)
(10)
The encoder is trained with specific unlearning objectives
defined on the forget and retain sets separately. Generally,
the unlearning objective for retain samples is to maintain the
base model’s predictive utility. We
leverage Preference Optimization (PO) (Maini et al. 2024)
to realize this objective in CAGUL by aligning the language
model’s outputs for forget queries with refusal responses instead of the ground truth. We compute the standard causal
language modeling loss with the substituted response as
Lf = E(V,X,Y )∼Df EY¯ t
log pθ(Y¯ t
|V, X, Y¯ <t)
(12)
where Y¯ is a refusal response such as “I cannot answer
this question. ” Finally, we formulate the joint objective for
CAGUL as
L = Lbce + Lf + Lr (13)Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Performance and Efficiency
Experimental results show that CAGUL performs better or on par with finetuning-based baselines without altering the pre-trained model parameters or incurring retraining costs. For LLaMA-3. p. 1 Formally, we can define the attention weights as a function of the query and key matrices:
A = softmax
QKT
√
d
(2)
In the context of VLMs, cross-modal attention is implemented as either cross-attention or joint self-attention. For
cross-attention-based models like LLaMA-3. CAGUL Rouge and EM scores on Df
demonstrate effective forgetting as noted by the significant
drops in these metrics compared to Finetune. The MIA score
MinK is also significantly lower than Finetune, which indicates that unlearned models achieved with CAGUL can successfully unlearn information from Df . We report retain metrics on the full
retain set Dr and the non-private subset Dnp corresponding
to the individuals in Df . We observe that CAGUL successfully preserves the retain utility of the finetuned base model. The baselines require fewer parameters than Retrain, as we implement LoRA finetuning for these methods
as opposed to the full finetuning used for Retrain. Nonetheless, CAGUL trains the fewest number of parameters as the
pre-trained VLM parameters are kept frozen in this framework and completes training in significantly less time.Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Ablation Insights
The ablation study indicates that CAGUL can achieve good trade-offs between forget and retain metrics without Cϕ, but its generalization ability is negatively affected. Conversely, CAGUL w/o Fψ shows a significant decline in forget and retain performance. p. 7 We note that the rouge score for Dr shows
significant improvement as the encoder is trained on the full
retain set, enabling the model to memorize additional knowledge. Conversely, we observe a significant decline in forget
and retain performance for CAGUL w/o Fψ, which shows
the importance of Fψ in our framework. The random selection of visual
tokens achieves mostly comparable performance for forget
metrics, but underperforms significantly for retain metrics. These results validate the selection of the least attended tokens for perturbation in CAGUL.Where the paper says it
Where the paper says it
Conclusion
In this work, we proposed a novel setting for unlearning in vision-language models where each user can exercise the Right to be Forgotten for any subset of their sensitive queries. We proposed Cross-Modal Attention Guided Unlearning (CAGUL), a framework consisting of a classifier to identify images appearing in the forget set, a cross-modal attention-based visual token selection mechanism, and a visual token encoder to embed unlearning information in visual tokens having low correlation with the query prompt. p. 8 8 Conclusion
In this work, we proposed a novel setting for unlearning
in vision-language models where each user can exercise
the Right to be Forgotten for any subset of their sensitive
queries. We proposed Cross-Modal Attention Guided Unlearning (CAGUL), a framework consisting of a classifier to
identify images appearing in the forget set, a cross-modal
attention-based visual token selection mechanism, and a visual token encoder to embed unlearning information in visual tokens having low correlation with the query prompt. We proposed Cross-Modal Attention Guided Unlearning (CAGUL), a framework consisting of a classifier to
identify images appearing in the forget set, a cross-modal
attention-based visual token selection mechanism, and a visual token encoder to embed unlearning information in visual tokens having low correlation with the query prompt. We conducted experiments on the FIUBench dataset using
two open-source vision-language models and show that our
method achieves desirable trade-offs in performance compared to finetuning-based baseline methods.Where the paper says it
Where the paper says it
References
Beniaguev, D. 2022. Synthetic Faces High Quality (SFHQ) dataset. p. 9 References
Beniaguev, D. Synthetic Faces High Quality (SFHQ)
dataset.Where the paper says it
Bhaila, K.; Van, M.-H.; and Wu, X. 2025. Soft Prompting for Unlearning in Large Language Models. p. 9 ; and Wu, X. Soft Prompting
for Unlearning in Large Language Models.Where the paper says it
Bourtoule, L. p. 0 In the LLM
landscape, research efforts have been made to mitigate privacy leakage in different ways, including machine unlearning. Specifically, machine unlearning addresses concerns regarding leakage of training data as stipulated by the California Consumer Privacy Act (CCPA) and GDPR’s Right to be
Forgotten (Cao and Yang 2015; Bourtoule et al. Bourtoule, L. ; Chandrasekaran, V. ; Chandrasekaran, V. ; Choquette-Choo, C. ; Choquette-Choo, C. ; Travers, A. However, training data collected from such
sources may contain personally identifiable and/or sensitive
information, raising significant privacy concerns, especially
with the added complexity of visual signals. For instance,
processing visual inputs in VLMs can unintentionally disclose information such as location cues, identity of individuals in the background, etc (Zhang et al. ; Zhang, B. ; and Papernot, N. However,
the models may memorize private and/or sensitive information during training and regurgitate it in inference. Recently,
machine unlearning has been leveraged to address the leakage of private data in LLMs.Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Cao, Y.; and Yang, J. p. 0 In the LLM
landscape, research efforts have been made to mitigate privacy leakage in different ways, including machine unlearning. Specifically, machine unlearning addresses concerns regarding leakage of training data as stipulated by the California Consumer Privacy Act (CCPA) and GDPR’s Right to be
Forgotten (Cao and Yang 2015; Bourtoule et al. ; and Yang, J. Towards making systems forget with machine unlearning.Where the paper says it
Where the paper says it
Chakraborty, T. p. 1 (2024b)
study how image-based manipulations can cause breaches
in safety-aligned VLMs. Chakraborty et al. Chakraborty, T. ; Shayegani, E. ; Shayegani, E. ; Abu-Ghazaleh,
N. ; Abu-Ghazaleh,
N. ; Roy-Chowdhury, A. ; Roy-Chowdhury, A. ; and Song,
C. ; and Song,
C. Can Textual Unlearning Solve Cross-Modality
Safety Alignment?Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Chen, J.; and Yang, D. p. 0 In the LLM
landscape, research efforts have been made to mitigate privacy leakage in different ways, including machine unlearning. Specifically, machine unlearning addresses concerns regarding leakage of training data as stipulated by the California Consumer Privacy Act (CCPA) and GDPR’s Right to be
Forgotten (Cao and Yang 2015; Bourtoule et al. ; and Yang, D. Unlearn What You Want to
Forget: Efficient Unlearning for LLMs.Where the paper says it
Where the paper says it
Cheng, J.; and Amiri, H. p. 1 VLM unlearning is a relatively new
frontier with increased complexity due to the integration of
visual signals. Cheng and Amiri (2023) propose a framework for multimodal unlearning in VLMs by decoupling
the text and visual modalities. ; and Amiri, H. Multimodal Machine Unlearning.Where the paper says it
Where the paper says it
Ding, C.; Wu, J.; Yuan, Y.; Lu, J.; Zhang, K. p. 0 However, training data collected from such
sources may contain personally identifiable and/or sensitive
information, raising significant privacy concerns, especially
with the added complexity of visual signals. For instance,
processing visual inputs in VLMs can unintentionally disclose information such as location cues, identity of individuals in the background, etc (Zhang et al. ; and He, X. Unified Parameter-Efficient Unlearning for LLMs.Where the paper says it
Where the paper says it
Faraglia, D. p. 4 1 Dataset
We primarily investigate VLM unlearning on the FIUBench
dataset (Ma et al. 2025) composed of 400 unique synthetic
images obtained from the SFHQ dataset (Beniaguev 2022),
each paired with fictitious biographical information including name, birthdate, address, phone number, occupation, income, health, and criminal records sourced from (Patil 2024;
Mendes 2020; Vyas 2017; Faraglia 2024).Where the paper says it
Fu, C.; Chen, P.; Shen, Y.; Qin, Y.; Zhang, M. p. 0 However, training data collected from such
sources may contain personally identifiable and/or sensitive
information, raising significant privacy concerns, especially
with the added complexity of visual signals. For instance,
processing visual inputs in VLMs can unintentionally disclose information such as location cues, identity of individuals in the background, etc (Zhang et al. ; Zhang, M. ; Zheng, X. ; and
Ji, R. MME: A Comprehensive Evaluation Benchmark for Multimodal Large Language Models.Where the paper says it
Where the paper says it
Where the paper says it
Gao, L.; Niu, Y.; Tang, T.; Avestimehr, S. p. 9 CoRR,
abs/2306. ; Avestimehr, S. ; Avestimehr, S. ; and Annavaram,
M. ; and Annavaram,
M. Ethos: Rectifying language models in orthogonal
parameter space.Where the paper says it
Where the paper says it
Where the paper says it
Gong, Y.; Ran, D.; Liu, J.; Wang, C.; Cong, T.; Wang, A.; Duan, S.; and Wang, X. p. 9 arXiv preprint arXiv:2403. ; and Wang, X. ; and Wang, X. FigStep: Jailbreaking Large
Vision-language Models via Typographic Visual Prompts.Where the paper says it
Where the paper says it
Kaduri, O.; Bagon, S. p. 0 Our method is motivated by the relationship between
visual and text tokens represented as cross-modal attention
scores computed between the two modes of input. Kaduri,
Bagon, and Dekel (2024) empirically demonstrated that
VLMs extract fine-grained details and attributes from visual
tokens in a spatially localized manner and compress visual
information into a small subset of highly attended tokens. Our method is motivated by the relationship between
visual and text tokens represented as cross-modal attention
scores computed between the two modes of input. Kaduri,
Bagon, and Dekel (2024) empirically demonstrated that
VLMs extract fine-grained details and attributes from visual
tokens in a spatially localized manner and compress visual
information into a small subset of highly attended tokens. Generally,
VLMs consist of a vision model that extracts visual features,
a cross-modal component that projects the visual features
to the LLM embedding space, and a language model that
processes projected visual tokens along with input text tokens for output generation. Together, these components enable VLMs to process both image and text modalities and be
utilized for a wide range of visio-linguistic tasks. What’s in the
Image? A Deep-Dive into the Vision of Vision Language
Models.Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Li, J.; Wei, Q.; Zhang, C. p. 0 However, training data collected from such
sources may contain personally identifiable and/or sensitive
information, raising significant privacy concerns, especially
with the added complexity of visual signals. For instance,
processing visual inputs in VLMs can unintentionally disclose information such as location cues, identity of individuals in the background, etc (Zhang et al. ;
and Liu, F. Single Image Unlearning: Efficient Machine Unlearning in Multimodal Large Language Models.Where the paper says it
Where the paper says it
Li, N.; Pan, A.; Gopal, A. p. 9 , Advances in Neural Information Processing Systems 38: Annual Conference
on Neural Information Processing Systems 2024, NeurIPS
2024, Vancouver, BC, Canada, December 10 - 15, 2024. ; Gopal, A. ; Gopal, A. ; Berrios, D. ; Berrios, D. ; Gatti, A. ; Gatti, A. ; Dombrowski, A. ; Dombrowski, A. The wmdp benchmark: Measuring and reducing malicious
use with unlearning.Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Liu, S.; Yao, Y.; Jia, J.; Casper, S. p. 9 Association for Computational Linguistics. ; Casper, S. ; Casper, S. ; Baracaldo, N. ; Baracaldo, N. Rethinking machine unlearning for large language models.Where the paper says it
Where the paper says it
Where the paper says it
Liu, X.; Zhu, Y.; Gu, J.; Lan, Y.; Yang, C.; and Qiao, Y. p. 9 arXiv
preprint arXiv:2402. ; and
Qiao, Y. ; and
Qiao, Y. MM-SafetyBench: A Benchmark for
Safety Evaluation of Multimodal Large Language Models.Where the paper says it
Where the paper says it
Liu, Z.; Dou, G.; Jia., M.; Tan, Z.; Zeng, Q.; Yuan, Y.; and Jiang, M. p. 9 arXiv:2311. ; and
Jiang, M. ; and
Jiang, M. Protecting Privacy in Multimodal Large
Language Models with MLLMU-Bench.Where the paper says it
Where the paper says it
Ma, Y.; Wang, J.; Wang, F.; Ma., S.; Li., J.; Pan., J.; Li., X.; Huang, F. p. 9 Association for Computational Linguistics. ;
Huang, F. ;
Huang, F. ; and Xiao,
C. ; and Xiao,
C. Benchmarking Vision Language Model Unlearning via Fictitious Facial Identity Dataset.Where the paper says it
Where the paper says it
Where the paper says it
Maini, P.; Feng, Z.; Schwarzschild, A. p. 9 In The Thirteenth
International Conference on Learning Representations. ; Schwarzschild, A. ; Schwarzschild, A. ; Lipton, Z. ; Lipton, Z. ; and
Kolter, J. ; and
Kolter, J. TOFU: A task of fictitious unlearning for
LLMs.Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Mendes, B., 2020. NYPD Crime Complaint Data Historic (2006-2019). p. 9 First Conference On Language Modeling. NYPD Crime Complaint Data Historic
(2006-2019).Where the paper says it
Patil, P., 2024. Healthcare Dataset. p. 9 NYPD Crime Complaint Data Historic
(2006-2019). Healthcare Dataset.Where the paper says it
Pawelczyk, M. p. 1 2024; Chen and Yang 2023; Zhang
et al. Others utilize adaptation techniques specific
to LLMs, such as achieving unlearning through in-context
learning (Pawelczyk, Neel, and Lakkaraju 2024), manually
crafted system prompts (Thaker, Maurya, and Smith 2024),
or prompt tuning with unlearning-specific losses to guide
generation (Bhaila, Van, and Wu 2025). 2024; Chen and Yang 2023; Zhang
et al. Others utilize adaptation techniques specific
to LLMs, such as achieving unlearning through in-context
learning (Pawelczyk, Neel, and Lakkaraju 2024), manually
crafted system prompts (Thaker, Maurya, and Smith 2024),
or prompt tuning with unlearning-specific losses to guide
generation (Bhaila, Van, and Wu 2025). ; and Lakkaraju, H. InContext Unlearning: Language Models as Few-Shot Unlearners.Where the paper says it
Where the paper says it
Where the paper says it
Samson, L.; Barazani, N. p. 9 In Forty-first International Conference on Machine
Learning, ICML 2024, Vienna, Austria, July 21-27, 2024. ; Barazani, N. ; Barazani, N. ; Ghebreab, S. ; Ghebreab, S. ; and Asano,
Y. ; and Asano,
Y. Privacy-Aware Visual Language Models.Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Shi, W.; Ajith, A. p. 9 arXiv:2405. ; Ajith, A. Association for Computational Linguistics. ;
Huang, F. ; Huang, Y. ; Blevins, T. ; Blevins, T. ; and Zettlemoyer, L. ; and Zettlemoyer, L. Detecting Pretraining
Data from Large Language Models.Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Thaker, P.; Maurya, Y. p. 1 2024; Chen and Yang 2023; Zhang
et al. Others utilize adaptation techniques specific
to LLMs, such as achieving unlearning through in-context
learning (Pawelczyk, Neel, and Lakkaraju 2024), manually
crafted system prompts (Thaker, Maurya, and Smith 2024),
or prompt tuning with unlearning-specific losses to guide
generation (Bhaila, Van, and Wu 2025). 2024; Chen and Yang 2023; Zhang
et al. Others utilize adaptation techniques specific
to LLMs, such as achieving unlearning through in-context
learning (Pawelczyk, Neel, and Lakkaraju 2024), manually
crafted system prompts (Thaker, Maurya, and Smith 2024),
or prompt tuning with unlearning-specific losses to guide
generation (Bhaila, Van, and Wu 2025). ; and Smith, V. Guardrail
baselines for unlearning in LLMs.Where the paper says it
Where the paper says it
Where the paper says it
Tomekc¸e, B. p. 1 There have also been several studies benchmarking VLM unlearning. Tomekc¸e et al. Tomekc¸e, B. ; Staab, R. ; Staab, R. ; and Vechev, M. ; and Vechev, M. Pri- ¨
vate Attribute Inference from Images with Vision-Language
Models.Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Vaswani, A. p. 1 We refer to the attention scores
between visual and text tokens as cross-modal attention. The
attention mechanism (Vaswani et al. For instance, a
user may want to remove references to sensitive information
(e. , social security number) but may not necessarily need
other general profile information to be removed (e.Where the paper says it
Where the paper says it
Vyas., 2017. USA Housing. p. 9 Protecting Privacy in Multimodal Large
Language Models with MLLMU-Bench. In Proceedings of
the 2025 Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics: Human Language Technologies, NAACL 2025 - Volume 1: Long
Papers, Albuquerque, New Mexico, USA, April 29 - May 4,
2025, 4105–4135.Where the paper says it
Yao, J.; Chien, E.; Du, M.; Niu, X.; Wang, T.; Cheng., Z.; and Yue., X. 2024. Machine Unlearning of Pre-trained Large Language Models. p. 9 ;
and Liu, F. Single Image Unlearning: Efficient Machine Unlearning in Multimodal Large Language Models.Where the paper says it
Yao., Y.; Xu., X., and Liu., Y. 2023. Large language model unlearning. p. 9 ; and Wu, X. Soft Prompting
for Unlearning in Large Language Models.Where the paper says it
Zhang, R.; Lin, L.; Bai, Y.; and Mei, S. 2024. Negative preference optimization: From catastrophic collapse to effective unlearning. p. 3 In CAGUL, we define this
objective as the standard causal language modeling loss for
for all samples in Dr, including the non-private samples for
images appearing in the forget set formulated as
Lr = E(V,X,Y )∼Dr EY t
log pθ(Y
t
|V, X, Y <t)
(11)
In the VQA task, the forget objective can be interpreted as
obtaining low utility for samples in the forget set Df . We
leverage Preference Optimization (PO) (Maini et al.Where the paper says it
Zhang., Z.; Bao, C.; Pan., X.; Chang, C.; Igarashi, T.; and Zhang, G. p. 0 However, training data collected from such
sources may contain personally identifiable and/or sensitive
information, raising significant privacy concerns, especially
with the added complexity of visual signals. For instance,
processing visual inputs in VLMs can unintentionally disclose information such as location cues, identity of individuals in the background, etc (Zhang et al. Our focus here is on VLM unlearning under practical privacy considerations. In VLMs, sensitive/private information
may be present in the vision component, language component, or both.Where the paper says it
Where the paper says it
--- Page 1 ---
Cross-Modal Attention Guided Unlearning in Vision-Language Models is a lightweight and efficient VLM unlearning framework that utilizes crossmodal attention to guide the encoding of unlearning information in visual tokens of low importance for relevant queries. It is practical and effective because it utilizes external modules to encode unlearning information in visual tokens of low importance for relevant queries. p. 0 We explore the role
of visual tokens for output generation in VLMs using crossmodal attention and utilize it to formulate Cross-Modal Attention Guided Unlearning (CAGUL), a lightweight and efficient VLM unlearning framework. In contrast to computationally expensive model finetuning methods, CAGUL utilizes external modules to encode unlearning information in
visual tokens of low importance for relevant queries.Where the paper says it
Pruning redundant reasoning tokens
The method leverages the relationship between visual and text tokens represented as cross-modal attention scores computed between the two modes of input. Kaduri, Bagon, and Dekel (2024) empirically demonstrated that VLMs extract fine-grained details and attributes from visual tokens in a spatially localized manner and compress visual information into a small subset of highly attended tokens. p. 0 Our method is motivated by the relationship between
visual and text tokens represented as cross-modal attention
scores computed between the two modes of input. Kaduri,
Bagon, and Dekel (2024) empirically demonstrated that
VLMs extract fine-grained details and attributes from visual
tokens in a spatially localized manner and compress visual
information into a small subset of highly attended tokens. that of prompting with all image tokens. Based on this observation, we conjecture that encoding information relevant
to private or non-private queries in the least attended visual
tokens can be useful for effective unlearning.Where the paper says it
Where the paper says it
How the training signal is built
The CAGUL framework trains two main components: a discriminator Cϕ and a visual token encoder Fψ. The discriminator is trained with the classification objective Lbce = EZv log pCϕ(lZv). p. 3 Our CAGUL framework has two
trainable modules, the discriminator Cϕ and the visual token
encoder Fψ. The discriminator is trained with the following
classification objective
Lbce = EZv
log pCϕ
(l|Zv)
(10)
The encoder is trained with specific unlearning objectives
defined on the forget and retain sets separately. Generally,
the unlearning objective for retain samples is to maintain the
base model’s predictive utility. In CAGUL, we define this
objective as the standard causal language modeling loss for
for all samples in Dr, including the non-private samples for
images appearing in the forget set formulated as
Lr = E(V,X,Y )∼Dr EY t
log pθ(Y
t
|V, X, Y <t)
(11)
In the VQA task, the forget objective can be interpreted as
obtaining low utility for samples in the forget set Df . In CAGUL, we define this
objective as the standard causal language modeling loss for
for all samples in Dr, including the non-private samples for
images appearing in the forget set formulated as
Lr = E(V,X,Y )∼Dr EY t
log pθ(Y
t
|V, X, Y <t)
(11)
In the VQA task, the forget objective can be interpreted as
obtaining low utility for samples in the forget set Df . We
leverage Preference Optimization (PO) (Maini et al.Where the paper says it
Where the paper says it
Where the paper says it
Where the gains come from
The primary gains are observed in forgetting quality, where CAGUL Rouge and EM scores on Df demonstrate effective forgetting compared to Finetune. Additionally, CAGUL successfully preserves the retain utility of the finetuned base model. p. 6 We report retain metrics on the full
retain set Dr and the non-private subset Dnp corresponding
to the individuals in Df . We observe that CAGUL successfully preserves the retain utility of the finetuned base model. Our results show that CAGUL retains the performance of
the finetuned base model on both MME and POPE datasets,
whereas finetuning baselines significantly degrade in performance. We emphasize that our method does not require finetuning any component of the VLM, which ensures that the
pretrained knowledge for general tasks is not manipulated,
provided that the classifier can accurately distinguish images
in the forget set from retain or general evaluation sets. 5-VL-7B-Instruct. Overall, our empirical results show that
CAGUL achieves desirable trade-offs between forget quality
and utility comparable to baseline finetuning methods while
saving on computational efficiency.Where the paper says it
Where the paper says it
Where the paper says it
Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.
Improvements for AI systems
-
Cross-Modal Attention Guided Unlearning (CAGUL) allows for unlearning in vision-language models by utilizing a discriminator to identify images in the forget set and a visual token encoder to embed unlearning information in visual tokens having low correlation with the query prompt. This enables removal of sensitive information without altering pre-trained model parameters or incurring retraining costs.
-
The system can perform subset forgetting, where
a user may want to forget a subset of their attributes available to the model
rather than forgetting all training data, as formulated in the problem setup for VLM unlearning. -
The framework can leverage cross-modal attention scores between visual and text modalities to select k least attended visual tokens with respect to the input text, guiding where unlearning objectives are encoded. This suggests that
encoding information relevant to private or non-private queries in the least attended visual tokens can be useful for effective unlearning.
-
The AI system can distinguish between forget set and retain set images using a discriminator module Cphi, which determines whether a given visual input is paired with a query about private data, formalized as l = Cphi(Zv).
-
The system can generate refusal responses for forget queries by leveraging Preference Optimization (PO) to align the language model’s outputs with
refusal responses instead of the ground truth
when training on the forget set Df. -
The unlearning objective is formulated as a joint loss L = Lbce + Lf + Lr, balancing binary classification loss for the discriminator, preference optimization loss for forget queries, and standard causal language modeling loss for retain samples.
Abstract
Vision-Language Models (VLMs) have demonstrated remarkable capabilities in multi-modal understanding for tasks such as Visual Question Answering (VQA), which requires models to infer outputs based on visual and textual context simultaneously. However, models may memorize some private and/or sensitive information during training and regurgitate it in inference. Recently, machine unlearning has been leveraged to address such leakage of private data in LLMs. VLMs add a layer of complexity to this process, as the visual context in the query may additionally contain sensitive information. To address this issue, we explore unlearning for VLMs, specifically for the VQA task. We explore the role of visual tokens for output generation in VLMs using cross-modal attention and utilize it to formulate Cross-Modal Attention Guided Unlearning (CAGUL), a lightweight and efficient VLM unlearning framework. In contrast to computationally expensive finetuning methods, CAGUL utilizes external modules to encode unlearning information in visual tokens of low importance for relevant queries. Empirical results show that the transformed visual tokens prevent leakage and retain reference model behavior to achieve desirable trade-offs in forget quality and model utility without altering pretrained model parameters.
Sources
- MultiDelete for Multimodal Machine Unlearning
- MME: A Comprehensive Evaluation Benchmark for Multimodal Large Language Models
- Ethos: Rectifying Language Models in Orthogonal Parameter Space
- FigStep: Jailbreaking Large Vision-Language Models via Typographic Visual Prompts
- What's in the Image? A Deep-Dive into the Vision of Vision Language Models
- The WMDP Benchmark: Measuring and Reducing Malicious Use With Unlearning
- Rethinking Machine Unlearning for Large Language Models
- MM-SafetyBench: A Benchmark for Safety Evaluation of Multimodal Large Language Models
- Privacy-Aware Visual Language Models
- Detecting Pretraining Data from Large Language Models
- Guardrail Baselines for Unlearning in LLMs
- Private Attribute Inference from Images with Vision-Language Models
- Large Language Model Unlearning
Related papers
- TEA: Text Encoder Alignment for Robust Concept Erasure in Text-to-Image Models
- Do Egocentric Video-Language Models Capture Both Hand- and Object-Centric Cues?
- Aura: Consistent Multi-Subject Video Generation via VLM-Grounded Semantic Alignment
- PVRF: All-in-one Adverse Weather Removal via Prior-modulated and Velocity-constrained Rectified Flow
- DLWM: Diverse Latent World Models for Efficient Multimodal Reasoning
- Out of Sight, Still in Mind: Token Compression for Omni-LLMs