Tag
This paper introduces ASAM (Adversarial Subspace Alignment for Multimodal Knowledge Editing), a method that improves the generality of intrinsic multimodal knowledge editing by generating adversarial yet semantically coherent variants and enforcing low-rank alignment of representations, addressing the limited generalization of existing methods.
This paper empirically evaluates vector merging methods for multilingual knowledge editing in large language models, identifying vector summation with shared covariance as the most reliable strategy and highlighting the limited effectiveness of Task Singular Vectors for Merging (TSVM) in reducing multilingual interference.
This paper identifies a failure mode called Entity Identity Confusion in multimodal knowledge editing, where models incorrectly bind image-entity relationships. It introduces EC-Bench to diagnose this issue and proposes mitigation strategies for faithful editing.