Research map / Adversarial robustness and security / Attacks on text and graphs
GeoDetect: Geometric Adversarial Detection for VLPs
Afsaneh Hasanebrahimi, Hanxun Huang, Christopher Leckie, James Bailey, Sarah Erfani
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use geometric scores (kNN) instead of MCM | CLIP-CNN · Zero-shot ImageNet1k, Sep_uni | AUC (%): 86.10 -> 99.65 | Improved |
| use geometric scores (LID) instead of MCM | CLIP-CNN · Zero-shot CIFAR100, Co-Attack | AUC (%): 43.99 -> 100 | Improved |
| use geometric scores (Mahalanobis) instead of MCM on multimodal attack | ALBEF · Zero-shot CIFAR10, Sep_multi | AUC (%): 47.43 -> 100 | Improved |
| kNN/KDE against multimodal-target attack on fused VLP | GeoDetect-kNN · Zero-shot ImageNet1k, ALBEF Sep_multi | AUC (%): 98.60 -> 51.92 | Hurt |
| add multimodal-encoder LID features | GeoDetect-LID · Zero-shot ImageNet1k, ALBEF Sep_multi | AUC (%): 51.92 -> 78.77 | Improved |
| kNN vs LID on retrieval, Sep_multi | GeoDetect · Retrieval Flickr30k, TCL Sep_multi | AUC (%): kNN 32.89 vs LID 84.72 | Hurt |
| adaptive attack against kNN detector (ablation) | GeoDetect-kNN · Retrieval Flickr30k, CLIP-CNN Sep_uni | AUC (%): 99.99 -> 67.82 | Hurt |
| adaptive attack against kNN detector (ablation) | GeoDetect-kNN · Retrieval Flickr30k, ALBEF Sep_uni | AUC (%): 99.75 -> 51.38 | Hurt |
| adaptive attack against LID detector (ablation) | GeoDetect-LID · Retrieval Flickr30k, CLIP-ViT Sep_uni | AUC (%): 99.37 -> 31.56 | Hurt |
| adaptive attack against LID detector (ablation) | GeoDetect-LID · Retrieval COCO, CLIP-CNN Sep_uni | AUC (%): 99.54 -> 93.14 | Hurt |
| adaptive attack against LID detector on multimodal-target attack (ablation) | GeoDetect-LID · Retrieval Flickr30k, ALBEF Sep_multi | AUC (%): 74.26 -> 89.83 | Improved |
Closest papers by meaning
- Adversarial Attacks Already Tell the Answer: Directional Bias-Guided Test-time Defense for Vision-Language Models (ICLR 2026 · similarity 0.87)
- The Adversarial Gait: Detecting Visual Adversarial Attacks against Vision-Language Models via Self-Targeted Gradient Characterization (NeurIPS 2026 · similarity 0.87)
- Visual Adversarial Examples Jailbreak Aligned Large Language Models (AAAI 2024 · similarity 0.86)
- Exploring Visual Vulnerabilities via Multi-Loss Adversarial Search for Jailbreaking Vision-Language Models (CVPR 2025 · similarity 0.86)
- Topological Signatures of Adversaries in Multimodal Alignments (ICML 2025 · similarity 0.85)
- PatchGuard: Adversarially Robust Anomaly Detection and Localization through Vision Transformers and Pseudo Anomalies (CVPR 2025 · similarity 0.85)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
