# Sparse Neuron Ablation Triggers Catastrophic Collapse of the Language Core in Large Vision-Language Models

arXiv: https://arxiv.org/abs/2512.00918 · HTML: https://mlsanity.com/paper/2512.00918/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Placed in Visual token pruning and efficiency, one of 16 directions in Multimodal LLMs / vision-language models.
- Builds on 18 of its 32 references that are in the map.
- First posted on arXiv 30 Nov 2025.

## Placed in

- Multimodal LLMs / vision-language models / [Visual token pruning and efficiency](https://mlsanity.com/mllm/visual-token-pruning-and-efficiency/)
- Mechanistic interpretability / [Circuits and mechanistic analysis](https://mlsanity.com/mechinterp/circuits-and-mechanistic-analysis/)

## Builds on (most cited first, 18 of 18)

- [Learning Transferable Visual Models From Natural Language Supervision](https://mlsanity.com/paper/2103.00020/) (2021)
- [Visual Instruction Tuning](https://mlsanity.com/paper/2304.08485/) (2023)
- [Improved Baselines with Visual Instruction Tuning](https://mlsanity.com/paper/2310.03744/) (2023)
- [InstructBLIP: Towards General-purpose Vision-Language Models with Instruction Tuning](https://mlsanity.com/paper/2305.06500/) (2023)
- [CLIPScore: A Reference-free Evaluation Metric for Image Captioning](https://mlsanity.com/paper/2104.08718/) (2021)
- [Deep Inside Convolutional Networks: Visualising Image Classification Models and Saliency Maps](https://mlsanity.com/paper/1312.6034/) (2013)
- [SNIP: Single-shot Network Pruning based on Connection Sensitivity](https://mlsanity.com/paper/1810.02340/) (2018)
- [When and why vision-language models behave like bags-of-words, and what to do about it?](https://mlsanity.com/paper/2210.01936/) (2022)
- [Mind the Gap: Understanding the Modality Gap in Multi-modal Contrastive Representation Learning](https://mlsanity.com/paper/2203.02053/) (2022)
- [Language-Specific Neurons: The Key to Multilingual Capabilities in Large Language Models](https://mlsanity.com/paper/2402.16438/) (2024)
- [On the Role of Attention Heads in Large Language Model Safety](https://mlsanity.com/paper/2410.13708/) (2024)
- [Ablation Studies in Artificial Neural Networks](https://mlsanity.com/paper/1901.08644/) (2019)
- [Confidence Regulation Neurons in Language Models](https://mlsanity.com/paper/2406.16254/) (2024)
- [The LLM Language Network: A Neuroscientific Approach for Identifying Causally Task-Relevant Units](https://mlsanity.com/paper/2411.02280/) (2024)
- [Optimal ablation for interpretability](https://mlsanity.com/paper/2409.09951/) (2024)
- [The Achilles' Heel of LLMs: How Altering a Handful of Neurons Can Cripple Language Abilities](https://mlsanity.com/paper/2510.10238/) (2025)
- [Neural Erosion: Emulating Controlled Neurodegeneration and Aging in AI Systems](https://mlsanity.com/paper/2403.10596/) (2024)
- [Dissecting Role Cognition in Medical LLMs via Neuronal Ablation](https://mlsanity.com/paper/2510.24677/) (2025)

## Abstract (arXiv)

Large Vision-Language Models (LVLMs) have shown impressive multimodal understanding capabilities, yet the structures that sustain their functionality remain poorly understood from a mechanistic interpretability standpoint. We propose Consistently Activated Neurons (CAN), a progressive neuron ablation method to identify critical neurons whose removal triggers catastrophic collapse, and use it to investigate structural vulnerabilities in representative 7B LVLMs. Experiments reveal that catastrophic collapse can be triggered by ablating as few as four neurons in textttLLaVA-1.5-7b-hf and a few thousand in textttInstructBLIP-vicuna-7b, both representing a small fraction of model parameters. Notably, critical neurons are predominantly localized in the language model, particularly in its down-projection layer, rather than in the vision components. We also observe a consistent two-stage collapse pattern: initial expressive degradation followed by sudden, complete collapse. These findings reveal that LVLM functionality depends on a sparse subset of neurons concentrated in the language backbone, offering mechanistic insights into how their functionality is structured and where these models are most vulnerable.