PRMU: A Corpus-Free Benchmark for Person-Centric Knowledge Unlearning in Multimodal Large Language Models
2026-08-11 • Computer Vision and Pattern Recognition
Computer Vision and Pattern Recognition
AI summaryⓘ
The authors created PRMU, a new way to test if big AI models that understand both pictures and words can forget specific person-related information without needing access to the original training data. They also developed a method called SGPE that gently edits the model's knowledge to remove targeted info while keeping related knowledge intact. Their experiments show that past methods struggle to forget with precision and avoid messing up nearby knowledge, but SGPE manages a better balance. The authors hope their work will help improve realistic and efficient ways to make AI models forget information when needed.
Multimodal Large Language ModelsMachine UnlearningPerson-centric Knowledge RemovalCorpus-free UnlearningKnowledge DisplacementParameter-space EditingLocality PreservationAdversarial EvaluationMultimodal Knowledge Reactivation
Authors
Huafeng Chen, Yueming Lyu, Ziyuan Chen, Wenda Tan, Chenyang Si, Liucheng Guo, Caifeng Shan
Abstract
Multimodal large language models (MLLMs) have demonstrated remarkable capabilities in storing and recalling rich person-related knowledge, raising increasing concerns about reliable knowledge removal. However, existing machine unlearning approaches for MLLMs typically assume access to original forget and retain corpora, which are often unavailable in realistic deletion scenarios. To address this limitation, we introduce PRMU, a benchmark for evaluating corpus-free multimodal unlearning under realistic person-centric deletion requests. PRMU focuses on naturally acquired person-related knowledge and evaluates whether models can remove target knowledge while preserving related knowledge through diverse textual and visual probes, including adversarial evaluation and fine-grained locality analysis. To facilitate research in this setting, we further introduce Similarity-Gated Projection Editing (SGPE), a lightweight corpus-free unlearning baseline with knowledge displacement, protected parameter-space editing, and locality-aware multimodal control. Extensive experiments on representative MLLMs reveal that existing unlearning methods often suffer from unfavorable forgetting-locality trade-offs, with significant locality degradation under aggressive forgetting settings, and remain vulnerable to multimodal knowledge reactivation. Meanwhile, SGPE provides a competitive trade-off between target forgetting, locality preservation, and general multimodal utility. We hope PRMU can facilitate future research toward realistic and scalable multimodal machine unlearning. Code and dataset will be released at https://github.com/2231122/PRMU.