Attribution Analysis Meets Model Editing: Advancing Knowledge Correction in Vision Language Models with VisEdit
This addresses the need for efficient knowledge correction in VLLMs, which is an incremental advancement as it adapts existing LLM editing techniques to the vision-language domain.
The paper tackles the problem of correcting outdated or erroneous knowledge in Vision-Language Models (VLLMs) without retraining, by proposing VisEdit, a model editor that edits intermediate visual representations based on attribution analysis, and shows superiority over strong baselines in benchmark evaluations.
Model editing aims to correct outdated or erroneous knowledge in large models without costly retraining. Recent research discovered that the mid-layer representation of the subject's final token in a prompt has a strong influence on factual predictions, and developed Large Language Model (LLM) editing techniques based on this observation. However, for Vision-LLMs (VLLMs), how visual representations impact the predictions from a decoder-only language model remains largely unexplored. To the best of our knowledge, model editing for VLLMs has not been extensively studied in the literature. In this work, we employ the contribution allocation and noise perturbation methods to measure the contributions of visual representations for token predictions. Our attribution analysis shows that visual representations in mid-to-later layers that are highly relevant to the prompt contribute significantly to predictions. Based on these insights, we propose VisEdit, a novel model editor for VLLMs that effectively corrects knowledge by editing intermediate visual representations in regions important to the edit prompt. We evaluated VisEdit using multiple VLLM backbones and public VLLM editing benchmark datasets. The results show the superiority of VisEdit over the strong baselines adapted from existing state-of-the-art editors for LLMs.