Minimal neuron ablation triggers catastrophic collapse in the language core of Large Vision-Language Models
This identifies critical safety risks in LVLMs, which is important for robustness and safety research, though it is incremental as it builds on existing vulnerability studies.
The paper investigated structural vulnerabilities in Large Vision-Language Models (LVLMs) and found that ablating as few as four neurons in the language model's feed-forward networks can trigger catastrophic collapse, with critical neurons localized in the language components and the down-projection layer being particularly vulnerable.
Large Vision-Language Models (LVLMs) have shown impressive multimodal understanding capabilities, yet their robustness is poorly understood. In this paper, we investigate the structural vulnerabilities of LVLMs to identify any critical neurons whose removal triggers catastrophic collapse. In this context, we propose CAN, a method to detect Consistently Activated Neurons and to locate critical neurons by progressive masking. Experiments on LLaVA-1.5-7b-hf and InstructBLIP-Vicuna-7b reveal that masking only a tiny portion of the language model's feed-forward networks (just as few as four neurons in extreme cases) suffices to trigger catastrophic collapse. Notably, critical neurons are predominantly localized in the language model rather than in the vision components, and the down-projection layer is a particularly vulnerable structure. We also observe a consistent two-stage collapse pattern: initial expressive degradation followed by sudden, complete collapse. Our findings provide important insights for safety research in LVLMs.