What Makes VLMs Robust? Towards Reconciling Robustness and Accuracy in Vision-Language Models

• URL: http://arxiv.org/abs/2603.12799v1
• Date: Fri, 13 Mar 2026 09:02:11 GMT
• Title: What Makes VLMs Robust? Towards Reconciling Robustness and Accuracy in Vision-Language Models
• Abstract summary: Adversarial Robustness Adaptation (R-Adapt) is a framework that freezes all pre-trained weights and introduces minimal, insight-driven adaptations only in the initial layers.<n>R-Adapt generalizes efficiently to large vision-language models (e.g., LLaVA and Qwen-VL) to enhance their robustness.
• Score: 64.99746027349767
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Achieving adversarial robustness in Vision-Language Models (VLMs) inevitably compromises accuracy on clean data, presenting a long-standing and challenging trade-off. In this work, we revisit this trade-off by investigating a fundamental question: What makes VLMs robust? Through a detailed analysis of adversarially fine-tuned models, we examine how robustness mechanisms function internally and how they interact with clean accuracy. Our analysis reveals that adversarial robustness is not uniformly distributed across network depth. Instead, unexpectedly, it is primarily localized within the shallow layers, driven by a low-frequency spectral bias and input-insensitive attention patterns. Meanwhile, updates to the deep layers tend to undermine both clean accuracy and robust generalization. Motivated by these insights, we propose Adversarial Robustness Adaptation (R-Adapt), a simple yet effective framework that freezes all pre-trained weights and introduces minimal, insight-driven adaptations only in the initial layers. This design achieves an exceptional balance between adversarial robustness and clean accuracy. R-Adapt further supports training-free, model-guided, and data-driven paradigms, offering flexible pathways to seamlessly equip standard models with robustness. Extensive evaluations on 18 datasets and diverse tasks demonstrate our state-of-the-art performance under various attacks. Notably, R-Adapt generalizes efficiently to large vision-language models (e.g., LLaVA and Qwen-VL) to enhance their robustness. Our project page is available at https://summu77.github.io/R-Adapt.

Related papers

• Robustifying Vision-Language Models via Test-Time Prompt Adaptation [23.05168102474528]
  We propose RITA, a test-tIme prompt-TAdaptation framework that shifts from sample-level estimates to distribution-level alignment.<n>Specifically, RITA employs optimal transport to align the distribution of augmented visual features with textual prototypes, mitigating adversarial outliers and rectifying cross-modal semantic misalignment.
  arXiv  Detail & Related papers  (2026-07-10T14:19:42Z)
• On Robustness and Chain-of-Thought Consistency of RL-Finetuned VLMs [15.301640007799735]
  We show that simple, controlled textual perturbations--misleading captions or incorrect chain-of-thought (CoT) traces--cause substantial drops in robustness and confidence.<n>To better understand these vulnerabilities, we analyze RL fine-tuning dynamics and uncover an accuracy-faithfulness trade-off.
  arXiv  Detail & Related papers  (2026-02-13T01:12:00Z)
• Same Answer, Different Representations: Hidden instability in VLMs [65.36933543377346]
  We introduce a representation-aware and frequency-aware evaluation framework that measures internal embedding drift, spectral sensitivity, and structural smoothness.<n>We apply this framework to modern Vision Language Models (VLMs) across the SEEDBench, MMMU, and POPE datasets.
  arXiv  Detail & Related papers  (2026-02-06T12:24:26Z)
• RoHOI: Robustness Benchmark for Human-Object Interaction Detection [84.78366452133514]
  Human-Object Interaction (HOI) detection is crucial for robot-human assistance, enabling context-aware support.<n>We introduce the first benchmark for HOI detection, evaluating model resilience under diverse challenges.<n>Our benchmark, RoHOI, includes 20 corruption types based on the HICO-DET and V-COCO datasets and a new robustness-focused metric.
  arXiv  Detail & Related papers  (2025-07-12T01:58:04Z)
• Sycophancy in Vision-Language Models: A Systematic Analysis and an Inference-Time Mitigation Framework [18.54098084470481]
  We analyze sycophancy across vision-language benchmarks and propose an inference-time mitigation framework.<n>Our framework effectively mitigates sycophancy across all evaluated models, while maintaining performance on neutral prompts.
  arXiv  Detail & Related papers  (2024-08-21T01:03:21Z)
• Efficient and Versatile Robust Fine-Tuning of Zero-shot Models [34.27380518351181]
  We introduce Robust Adapter (R-Adapter), a novel method for fine-tuning zero-shot models to downstream tasks. Our method integrates lightweight modules into the pre-trained model and employs novel self-ensemble techniques to boost OOD robustness and reduce storage expenses substantially. Our experiments demonstrate that R-Adapter achieves state-of-the-art performance across a diverse set of tasks, tuning only 13% of the parameters of the CLIP encoders.
  arXiv  Detail & Related papers  (2024-08-11T11:37:43Z)
• Adversarial Robustification via Text-to-Image Diffusion Models [56.37291240867549]
  Adrial robustness has been conventionally believed as a challenging property to encode for neural networks. We develop a scalable and model-agnostic solution to achieve adversarial robustness without using any data.
  arXiv  Detail & Related papers  (2024-07-26T10:49:14Z)
• Towards a robust and reliable deep learning approach for detection of compact binary mergers in gravitational wave data [0.0]
  We develop a deep learning model stage-wise and work towards improving its robustness and reliability. We retrain the model in a novel framework involving a generative adversarial network (GAN) Although absolute robustness is practically impossible to achieve, we demonstrate some fundamental improvements earned through such training.
  arXiv  Detail & Related papers  (2023-06-20T18:00:05Z)
• Self-Damaging Contrastive Learning [92.34124578823977]
  Unlabeled data in reality is commonly imbalanced and shows a long-tail distribution. This paper proposes a principled framework called Self-Damaging Contrastive Learning to automatically balance the representation learning without knowing the classes. Our experiments show that SDCLR significantly improves not only overall accuracies but also balancedness.
  arXiv  Detail & Related papers  (2021-06-06T00:04:49Z)
• RobustBench: a standardized adversarial robustness benchmark [84.50044645539305]
  Key challenge in benchmarking robustness is that its evaluation is often error-prone leading to robustness overestimation. We evaluate adversarial robustness with AutoAttack, an ensemble of white- and black-box attacks. We analyze the impact of robustness on the performance on distribution shifts, calibration, out-of-distribution detection, fairness, privacy leakage, smoothness, and transferability.
  arXiv  Detail & Related papers  (2020-10-19T17:06:18Z)

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.