Towards Reasoning-Preserving Unlearning in Multimodal Large Language Models

Fuente: arXiv
Guardado en:
Detalles Bibliográficos
Autores principales: Li, Hongji, yao, Junchi, Yu, Manjiang, Singh, Priyanka, Li, Xue, Wang, Di, Hu, Lijie
Formato: Preprint
Publicado: 2025
Materias:
Acceso en línea:
Etiquetas: Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
_version_ 1866917159408500736
author Li, Hongji
yao, Junchi
Yu, Manjiang
Singh, Priyanka
Li, Xue
Wang, Di
Hu, Lijie
author_facet Li, Hongji
yao, Junchi
Yu, Manjiang
Singh, Priyanka
Li, Xue
Wang, Di
Hu, Lijie
contents Machine unlearning aims to erase requested data from trained models without full retraining. For Reasoning Multimodal Large Language Models (RMLLMs), this is uniquely challenging: intermediate chain-of-thought steps can still leak sensitive information even when final answers are forgotten, and overly aggressive interventions easily damage general reasoning ability. Yet no benchmark jointly evaluates how well unlearning methods suppress reasoning-level leakage while preserving reasoning competence. We address this gap with RMLLMU-Bench, the first benchmark for RMLLM unlearning that extends standard forgetting metrics with dedicated measures of reasoning leakage and reasoning retention. A systematic evaluation on RMLLMU-Bench reveals that existing unlearning methods for MLLMs and Large (Language) Reasoning Models (LRMs) either leave substantial leakage in the reasoning process or severely degrade reasoning performance. To address these gaps, we propose R-MUSE (Reasoning-preserving MLLM Unlearning via Subspace guidance and Adaptive Steering), a training-free and inference-time intervention framework that steers internal representations to forget both answers and reasoning traces while explicitly preserving general reasoning. Experiments on RMLLMU-Bench demonstrate that R-MUSE achieves a substantially better balance between effective forgetting and reasoning retention.
format Preprint
id arxiv_https___arxiv_org_abs_2512_17911
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Towards Reasoning-Preserving Unlearning in Multimodal Large Language Models
Li, Hongji
yao, Junchi
Yu, Manjiang
Singh, Priyanka
Li, Xue
Wang, Di
Hu, Lijie
Computation and Language
Artificial Intelligence
Machine Learning
Machine unlearning aims to erase requested data from trained models without full retraining. For Reasoning Multimodal Large Language Models (RMLLMs), this is uniquely challenging: intermediate chain-of-thought steps can still leak sensitive information even when final answers are forgotten, and overly aggressive interventions easily damage general reasoning ability. Yet no benchmark jointly evaluates how well unlearning methods suppress reasoning-level leakage while preserving reasoning competence. We address this gap with RMLLMU-Bench, the first benchmark for RMLLM unlearning that extends standard forgetting metrics with dedicated measures of reasoning leakage and reasoning retention. A systematic evaluation on RMLLMU-Bench reveals that existing unlearning methods for MLLMs and Large (Language) Reasoning Models (LRMs) either leave substantial leakage in the reasoning process or severely degrade reasoning performance. To address these gaps, we propose R-MUSE (Reasoning-preserving MLLM Unlearning via Subspace guidance and Adaptive Steering), a training-free and inference-time intervention framework that steers internal representations to forget both answers and reasoning traces while explicitly preserving general reasoning. Experiments on RMLLMU-Bench demonstrate that R-MUSE achieves a substantially better balance between effective forgetting and reasoning retention.
title Towards Reasoning-Preserving Unlearning in Multimodal Large Language Models
topic Computation and Language
Artificial Intelligence
Machine Learning
url https://arxiv.org/abs/2512.17911