Unraveling and Mitigating Safety Alignment Degradation of Vision-Language Models

• URL: http://arxiv.org/abs/2410.09047v1
• Date: Fri, 11 Oct 2024 17:59:31 GMT
• Title: Unraveling and Mitigating Safety Alignment Degradation of Vision-Language Models
• Authors: Qin Liu, Chao Shang, Ling Liu, Nikolaos Pappas, Jie Ma, Neha Anna John, Srikanth Doss, Lluis Marquez, Miguel Ballesteros, Yassine Benajiba,
• Abstract summary: The safety alignment ability of Vision-Language Models (VLMs) is prone to be degraded by the integration of the vision module. We show that the challenge arises from the representation gap that emerges when introducing vision modality to VLMs. To reduce safety alignment degradation, we introduce Cross-Modality Representation Manipulation (CMRM)
• Score: 26.83278034227966
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: The safety alignment ability of Vision-Language Models (VLMs) is prone to be degraded by the integration of the vision module compared to its LLM backbone. We investigate this phenomenon, dubbed as ''safety alignment degradation'' in this paper, and show that the challenge arises from the representation gap that emerges when introducing vision modality to VLMs. In particular, we show that the representations of multi-modal inputs shift away from that of text-only inputs which represent the distribution that the LLM backbone is optimized for. At the same time, the safety alignment capabilities, initially developed within the textual embedding space, do not successfully transfer to this new multi-modal representation space. To reduce safety alignment degradation, we introduce Cross-Modality Representation Manipulation (CMRM), an inference time representation intervention method for recovering the safety alignment ability that is inherent in the LLM backbone of VLMs, while simultaneously preserving the functional capabilities of VLMs. The empirical results show that our framework significantly recovers the alignment ability that is inherited from the LLM backbone with minimal impact on the fluency and linguistic capabilities of pre-trained VLMs even without additional training. Specifically, the unsafe rate of LLaVA-7B on multi-modal input can be reduced from 61.53% to as low as 3.15% with only inference-time intervention. WARNING: This paper contains examples of toxic or harmful language.

Related papers

• Preserving Multi-Modal Capabilities of Pre-trained VLMs for Improving Vision-Linguistic Compositionality [69.76121008898677]
  Fine-grained Selective Calibrated CLIP integrates local hard negative loss and selective calibrated regularization. Our evaluations show that FSC-CLIP not only achieves compositionality on par with state-of-the-art models but also retains strong multi-modal capabilities.
  arXiv  Detail & Related papers  (2024-10-07T17:16:20Z)
• CoCA: Regaining Safety-awareness of Multimodal Large Language Models with Constitutional Calibration [90.36429361299807]
  multimodal large language models (MLLMs) have demonstrated remarkable success in engaging in conversations involving visual inputs. The integration of visual modality has introduced a unique vulnerability: the MLLM becomes susceptible to malicious visual inputs. We introduce a technique termed CoCA, which amplifies the safety-awareness of the MLLM by calibrating its output distribution.
  arXiv  Detail & Related papers  (2024-09-17T17:14:41Z)
• What Do VLMs NOTICE? A Mechanistic Interpretability Pipeline for Gaussian-Noise-free Text-Image Corruption and Evaluation [16.033361754660316]
  Notice is the first Noise-free Text-Image Corruption and Evaluation pipeline for interpretability in Vision-Language Models (VLMs) Our experiments on the SVO-Probes, MIT-States, and Facial Expression Recognition datasets reveal crucial insights into VLM decision-making. This work paves the way for more transparent and interpretable multimodal systems.
  arXiv  Detail & Related papers  (2024-06-24T05:13:19Z)
• Cross-Modal Safety Alignment: Is textual unlearning all you need? [36.29740845754985]
  We show that unlearning solely in the textual domain can be effective for cross-modality safety alignment. Our experiments show that unlearning with a multi-modal dataset offers no potential benefits but incurs significantly increased computational demands.
  arXiv  Detail & Related papers  (2024-05-27T20:29:13Z)
• Safety Alignment for Vision Language Models [21.441662865727448]
  We enhance the visual modality safety alignment of Vision Language Models (VLMs) by adding safety modules. Our method boasts ease of use, high flexibility, and strong controllability, and it enhances safety while having minimal impact on the model's general performance.
  arXiv  Detail & Related papers  (2024-05-22T12:21:27Z)
• RigorLLM: Resilient Guardrails for Large Language Models against Undesired Content [62.685566387625975]
  Current mitigation strategies, while effective, are not resilient under adversarial attacks. This paper introduces Resilient Guardrails for Large Language Models (RigorLLM), a novel framework designed to efficiently moderate harmful and unsafe inputs.
  arXiv  Detail & Related papers  (2024-03-19T07:25:02Z)
• Eyes Closed, Safety On: Protecting Multimodal LLMs via Image-to-Text Transformation [98.02846901473697]
  We propose ECSO (Eyes Closed, Safety On), a training-free protecting approach that exploits the inherent safety awareness of MLLMs. ECSO generates safer responses via adaptively transforming unsafe images into texts to activate the intrinsic safety mechanism of pre-aligned LLMs.
  arXiv  Detail & Related papers  (2024-03-14T17:03:04Z)
• Machine Vision Therapy: Multimodal Large Language Models Can Enhance Visual Robustness via Denoising In-Context Learning [67.0609518552321]
  We propose to conduct Machine Vision Therapy which aims to rectify the noisy predictions from vision models. By fine-tuning with the denoised labels, the learning model performance can be boosted in an unsupervised manner.
  arXiv  Detail & Related papers  (2023-12-05T07:29:14Z)
• Visual Adversarial Examples Jailbreak Aligned Large Language Models [66.53468356460365]
  We show that the continuous and high-dimensional nature of the visual input makes it a weak link against adversarial attacks. We exploit visual adversarial examples to circumvent the safety guardrail of aligned LLMs with integrated vision. Our study underscores the escalating adversarial risks associated with the pursuit of multimodality.
  arXiv  Detail & Related papers  (2023-06-22T22:13:03Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.