TRU: Targeted Reverse Update for Efficient Multimodal Recommendation Unlearning

Dimensions overall score 7.0

GitHub Code Pulse

No public code linked for this paper yet.

• Existing approximate unlearning methods for multimodal recommendation systems rely on a largely uniform reverse update across the model.limitation 90%
• The influence of deleted data in modern multimodal recommendation systems is not uniformly distributed but is concentrated unevenly across ranking behavior, modality branches, and network layers.result 90%
• This non-uniform influence creates three bottlenecks in MRS unlearning: target-item persistence in the collaborative graph, modality imbalance across feature branches, and layer-wise sensitivity in the parameter space.technical 85%
• The proposed Targeted Reverse Update (TRU) framework performs three coordinated interventions: a ranking fusion gate, branch-wise modality scaling, and capacity-aware layer isolation.method 90%
• TRU consistently achieves a better retain-forget trade-off than prior approximate baselines across two backbones, three datasets, and three unlearning regimes.result 85%
• Security audits confirm that TRU achieves deeper forgetting and behavior closer to a full retraining on the retained data.result 80%
• TRU is a plug-and-play unlearning framework for multimodal recommendation systems.method 90%

Competitive landscape

Competitor map is still being generated for this paper. Enable generation or check back soon.

Startup potential card