Research map / Adversarial robustness and security / Attacks on text and graphs
Cross-Modal Learning with Adversarial Samples
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| apply CMLA adversarial perturbation to query (100 to 500 iterations) | DCMH · MIRFlickr-25K image query vs text database, 32 bits | MAP: 0.749 -> 0.521 | Hurt |
| apply CMLA adversarial perturbation to query (500 iterations) | SSAH · NUS-WIDE text query vs image database, 32 bits | MAP: 0.677 -> 0.351 | Hurt |
| increase attack iterations from 100 to 500 | CMLA · MIRFlickr-25K image query vs text database, DCMH | MAP and distortion D: MAP 0.579 -> 0.521, D 0.039 -> 0.019 | Helped |
| add intra-modal similarity loss to adversarial sample learning (ablation) | CMLA on SSAH · MIRFlickr-25K single-modal retrieval I2I | MAP: 0.420 -> 0.674 | Helped |
| add intra-modal similarity loss to adversarial sample learning (ablation) | CMLA on SSAH · MIRFlickr-25K cross-modal I2T/T2I attack strength | MAP drop: 15.7% and 9.3% larger drops without the loss | Hurt |
| adversarial training with CMLA samples | DCMH · MIRFlickr-25K image query vs text database, 32 bits | MAP under attack: 0.521 -> 0.723 | Helped |
| adversarial training with CMLA samples | SSAH · NUS-WIDE image query vs text database, 32 bits | MAP under attack: 0.460 -> 0.640 | Helped |
| replace vgg-f with ResNet50 backbone | DCMH · MIRFlickr-25K image query vs text database, 32 bits | MAP (clean): 0.749 -> 0.816 | Helped |
Closest papers by meaning
- Self-Supervised Adversarial Hashing Networks for Cross-Modal Retrieval (CVPR 2018 · similarity 0.90)
- Unsupervised Generative Adversarial Cross-Modal Hashing (AAAI 2018 · similarity 0.90)
- Mutual Quantization for Cross-Modal Search With Noisy Labels (CVPR 2022 · similarity 0.88)
- Dual Self-Paced Cross-Modal Hashing (AAAI 2024 · similarity 0.86)
- Robust Self-Paced Hashing for Cross-Modal Retrieval with Noisy Labels (AAAI 2025 · similarity 0.86)
- Lightweight Contrastive Distilled Hashing for Online Cross-modal Retrieval (AAAI 2025 · similarity 0.84)
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.