Beyond Sample-Level Forgetting: Improving Reliability in Multimodal Unlearning

Jianzhou Wang, Yirui Wu, Lixin Yuan, Wenxiao Zhang, Jun Liu
Proceedings of the 43rd International Conference on Machine Learning, PMLR 306:131263-131278, 2026.

Abstract

Multimodal unlearning aims to eliminate specific data from pretrained multimodal models, which offers significant advantages in data privacy and model efficiency. Current methods struggle to achieve the desired properties of effectiveness, reliability and locality, due to the complex interdependency of unimodal and multimodal knowledge. By introducing a causal perspective, we propose multimodal unlearning with decoupled knowledge components. To promote fine-grained understanding of multimodal context, we introduce Multimodal Variational Inference (MVI) to infer modal-specific and -consistent factors with incomplete sample observation. With foundation of decoupled knowledge, we propose contrastive semantic editing to regulate multimodal unlearning towards refined forgetting. Experiments on privacy- and copyright-sensitive scenarios validate effectiveness of our method across multiple scenarios, ensuring the unlearned model maintains high reliability and locality.

Cite this Paper


BibTeX
@InProceedings{pmlr-v306-wang26ja, title = {Beyond Sample-Level Forgetting: Improving Reliability in Multimodal Unlearning}, author = {Wang, Jianzhou and Wu, Yirui and Yuan, Lixin and Zhang, Wenxiao and Liu, Jun}, booktitle = {Proceedings of the 43rd International Conference on Machine Learning}, pages = {131263--131278}, year = {2026}, editor = {Zhang, Tong and Dudik, Miroslav and Jaggi, Martin and Agarwal, Alekh and Li, Sharon and Schuurmans, Dale and Zhu, Jerry and Berkenkamp, Felix and Dong, Hanze and Bietti, Alberto}, volume = {306}, series = {Proceedings of Machine Learning Research}, month = {06--11 Jul}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v306/main/assets/wang26ja/wang26ja.pdf}, url = {https://proceedings.mlr.press/v306/wang26ja.html}, abstract = {Multimodal unlearning aims to eliminate specific data from pretrained multimodal models, which offers significant advantages in data privacy and model efficiency. Current methods struggle to achieve the desired properties of effectiveness, reliability and locality, due to the complex interdependency of unimodal and multimodal knowledge. By introducing a causal perspective, we propose multimodal unlearning with decoupled knowledge components. To promote fine-grained understanding of multimodal context, we introduce Multimodal Variational Inference (MVI) to infer modal-specific and -consistent factors with incomplete sample observation. With foundation of decoupled knowledge, we propose contrastive semantic editing to regulate multimodal unlearning towards refined forgetting. Experiments on privacy- and copyright-sensitive scenarios validate effectiveness of our method across multiple scenarios, ensuring the unlearned model maintains high reliability and locality.} }
Endnote
%0 Conference Paper %T Beyond Sample-Level Forgetting: Improving Reliability in Multimodal Unlearning %A Jianzhou Wang %A Yirui Wu %A Lixin Yuan %A Wenxiao Zhang %A Jun Liu %B Proceedings of the 43rd International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2026 %E Tong Zhang %E Miroslav Dudik %E Martin Jaggi %E Alekh Agarwal %E Sharon Li %E Dale Schuurmans %E Jerry Zhu %E Felix Berkenkamp %E Hanze Dong %E Alberto Bietti %F pmlr-v306-wang26ja %I PMLR %P 131263--131278 %U https://proceedings.mlr.press/v306/wang26ja.html %V 306 %X Multimodal unlearning aims to eliminate specific data from pretrained multimodal models, which offers significant advantages in data privacy and model efficiency. Current methods struggle to achieve the desired properties of effectiveness, reliability and locality, due to the complex interdependency of unimodal and multimodal knowledge. By introducing a causal perspective, we propose multimodal unlearning with decoupled knowledge components. To promote fine-grained understanding of multimodal context, we introduce Multimodal Variational Inference (MVI) to infer modal-specific and -consistent factors with incomplete sample observation. With foundation of decoupled knowledge, we propose contrastive semantic editing to regulate multimodal unlearning towards refined forgetting. Experiments on privacy- and copyright-sensitive scenarios validate effectiveness of our method across multiple scenarios, ensuring the unlearned model maintains high reliability and locality.
APA
Wang, J., Wu, Y., Yuan, L., Zhang, W. & Liu, J.. (2026). Beyond Sample-Level Forgetting: Improving Reliability in Multimodal Unlearning. Proceedings of the 43rd International Conference on Machine Learning, in Proceedings of Machine Learning Research 306:131263-131278 Available from https://proceedings.mlr.press/v306/wang26ja.html.

Related Material