arXiv new/cross + ICML 2026 regular · P2 · 2026-07-14

Robustifying Vision-Language Models via Test-Time：它和通用视觉自监督的关系在于：测试时用分布级 alignment 强化 CLIP/VLM adversarial robustness，关注预训练视觉-语言表征的脆弱性

中高相关；详见方法、贡献和实验边界。

Robustifying Vision-Language Models via Test-Time Prompt Adaptation　arXiv new/cross + ICML 2026 regular　原文链接

编号2607.09450 优先级P2 类别arXiv new/cross + ICML 2026 regular 会议arXiv new/cross + ICML 2026 regular 方法测试时用分布级 alignment 强化 CLIP/VLM adversarial robustness，关注预训练视觉-语言表征的脆弱性来源arXiv / OpenReview

先说结论。它和通用视觉自监督的关系在于：测试时用分布级 alignment 强化 CLIP/VLM adversarial robustness，关注预训练视觉-语言表征的脆弱性。中高相关；详见方法、贡献和实验边界。

**Figureure 2 · Overview of the proposed RITA framework**Figure 2. Overview of the proposed RITA framework. Given an adversarial test image, RITA extracts multi-view visual features and class-specific textual prototypes using a frozen CLIP encoder. Both modalities are modeled as discrete distributions and aligned via entropy-regularized optimal transport. Low-entropy views are used to update a dynamic cache of reliable semantics.这张图概括 Robustifying Vision-Language Models via Test-Time 的整体方法流程。阅读时先看模块之间传递的训练信号，再看作者如何把目标拆成可优化的子问题。

**Figureure 1 · Augmented views retain more semantic cues under adversarial perturbati**Figure 1. Augmented views retain more semantic cues under adversarial perturbations, enabeling a cache for distribution alignment that improves adversarial performance. (a) Visualization of adversarially perturbed images, where each point represents an image and different colors denote ground-truth classes. (b) Visualization of multiple augmented views generated from the same adversarial image, colored by class label, where semantic structure partially re-emerges with improved class separability compared to (a). (c) Our method leverages the selected augmented views as a cache and aligns them with textual prompts. (d) Performance comparison across different VLM backbones, demonstrating improved robustness under adversarial attacks.这张图/表用于判断 Robustifying Vision-Language Models via Test-Time 的实验收益来自哪里。重点看替换、消融或跨模型设置下趋势是否一致，而不是只看单个最高分。

核心问题

它和通用视觉自监督的关系在于：测试时用分布级 alignment 强化 CLIP/VLM adversarial robustness，关注预训练视觉-语言表征的脆弱性。

方法拆解

测试时用分布级 alignment 强化 CLIP/VLM adversarial robustness，关注预训练视觉-语言表征的脆弱性

主要贡献

中高相关；详见方法、贡献和实验边界。

实验看点

实验部分建议重点看两类证据：一是作者是否把方法收益和更强数据、更长训练、更大模型区分开；二是跨模型、跨数据或跨任务迁移是否还能保留同样趋势。

局限与读法

这篇论文的结论需要结合任务设置、训练数据规模和消融实验一起看；不要只凭单个指标判断它对通用视觉表征的价值。