AIMay 30, 2025

Bootstrapping LLM Robustness for VLM Safety via Reducing the Pretraining Modality Gap

arXiv:2505.24208v14 citationsh-index: 29
Originality Incremental advance
AI Analysis

This addresses a critical safety issue for deploying VLMs in real-world applications, though it is an incremental improvement on existing models.

The paper tackles the problem of safety degradation in Vision-Language Models (VLMs) by showing that the modality gap between image and text representations is highly inversely correlated with safety, and proposes a regularization method during pretraining to reduce this gap, resulting in up to 16.3% reduction in unsafe rates without compromising performance.

Ensuring Vision-Language Models (VLMs) generate safe outputs is crucial for their reliable deployment. However, LVLMs suffer from drastic safety degradation compared to their LLM backbone. Even blank or irrelevant images can trigger LVLMs to generate harmful responses to prompts that would otherwise be refused in text-only contexts. The modality gap between image and text representations has been recently hypothesized to contribute to safety degradation of LVLMs. However, if and how the amount of modality gap affects LVLMs' safety is not studied. In this work, we show that the amount of modality gap is highly inversely correlated with VLMs' safety. Then, we show that this modality gap is introduced during pretraining LVLMs and persists through fine-tuning. Inspired by this observation, we propose a regularization to reduce the modality gap during pretraining. Our extensive experiments on LLaVA v1.5, ShareGPT4V, and MiniGPT-4 show that our method substantially improves safety alignment of LVLMs, reducing unsafe rate by up to 16.3% without compromising performance, and can further boost existing defenses by up to 18.2%.

Foundations

The foundational work for this paper's niche, ranked by how specifically the neighbourhood builds on it — not by global fame.

Your Notes