Research map / Adversarial robustness and security / Attacks on text and graphs
Adversarial Training with Fast Gradient Projection Method against Synonym Substitution Based Text Attacks
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 |
|---|---|---|---|
| Use FGPM's gradient-projection synonym selection instead of query-based GSA | text synonym-substitution attacks · AG's News, DBPedia, and Yahoo! Answers | attack-generation time: 8 vs 276 seconds (AG's News, CNN) | Helped |
| Use FGPM rather than IGA | text synonym-substitution attacks · AG's News, DBPedia, and Yahoo! Answers | classification accuracy under attack: 37.5 vs 30.0; 40.0 vs 36.5; 6.0 vs 3.5 | Mixed |
| Train with FGPM adversarial examples and logit pairing | CNN classifier · AG's News, FGPM attack | classification accuracy (%): 37.5 -> 88.0 | Helped |
| Use ATFL instead of SEM | CNN classifier · AG's News, clean test set | classification accuracy (%): 87.5 -> 89.0 | Helped |
| Use ATFL instead of SEM | CNN classifier · AG's News, FGPM attack | classification accuracy (%): 78.5 -> 88.0 | Helped |
| Vary alpha in ATFL while fixing lambda at 0.5 (ablation) | ATFL · AG's News, CNN, attacks | classification accuracy (%): 85.5 to 88.0 | No clear effect |
| Vary lambda in ATFL while fixing alpha at 0.5 (ablation) | ATFL · AG's News, CNN, attacks | classification accuracy (%): 85.5 to 88.0 | No clear effect |
| Use ALP rather than standard adversarial training (ablation) | CNN text classifier · AG's News, FGPM attack | classification accuracy (%): 78.0 -> 88.0 | Helped |
| Use TRADES rather than standard adversarial training (ablation) | CNN text classifier · AG's News, FGPM attack | classification accuracy (%): 78.0 -> 40.0 | Hurt |
| Use CLP rather than standard adversarial training (ablation) | CNN text classifier · AG's News, FGPM attack | classification accuracy (%): 78.0 -> 38.5 | Hurt |
Closest papers by meaning
- TextGrad: Advancing Robustness Evaluation in NLP by Gradient-Driven Optimization (ICLR 2023 · similarity 0.87)
- Bigram and Unigram Based Text Attack via Adaptive Monotonic Heuristic Search (AAAI 2021 · similarity 0.87)
- Taxonomy Driven Fast Adversarial Training (AAAI 2024 · similarity 0.86)
- Improved Generation of Adversarial Examples Against Safety-aligned LLMs (NeurIPS 2024 · similarity 0.86)
- Revisiting Character-level Adversarial Attacks for Language Models (ICML 2024 · similarity 0.84)
- Adversarial Paraphrasing: A Universal Attack for Humanizing AI-Generated Text (NeurIPS 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.