Research map / Adversarial robustness and security / Certified and empirical robustness
Greedy Attack and Gumbel Attack: Generating Adversarial Examples for Discrete Data
Puyudi Yang, Jianbo Chen, Cho-Jui Hsieh, Jane-Ling Wang, Michael I. Jordan
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 Greedy Attack instead of Delete-1, DeepWordBug, Saliency and projected FGSM baselines | word-based models (IMDB Word-CNN, Yahoo LSTM) · IMDB and Yahoo! Answers | attack success rate | Improved |
| use Gumbel Attack versus Saliency-projected FGSM | Yahoo! Answers LSTM · Yahoo! Answers | attack success rate | Hurt |
| use Greedy Attack on character-based CNN versus other methods | Char-CNN · AG's News | attack success rate | Improved |
| transfer Greedy Attack examples to other models | Greedy Attack · IMDB (transfer to CNN2, LSTM) | success rate | Hurt |
| transfer Gumbel Attack examples to other model structures | Gumbel Attack · IMDB and Yahoo! Answers | success rate | No clear effect |
| perturb 2 words in IMDB reviews: effect on humans versus models | Greedy and Gumbel Attack · IMDB human evaluation | accuracy drop: human drop 10.5% and 7.5% vs. model drop 75% and 25% | Mixed |
Closest papers by meaning
- Revisiting Character-level Adversarial Attacks for Language Models (ICML 2024 · similarity 0.85)
- Improved Generation of Adversarial Examples Against Safety-aligned LLMs (NeurIPS 2024 · similarity 0.84)
- Confidence Elicitation: A New Attack Vector for Large Language Models (ICLR 2025 · similarity 0.84)
- LimeAttack: Local Explainable Method for Textual Hard-Label Adversarial Attack (AAAI 2024 · similarity 0.84)
- Statistical Estimation of Adversarial Risk in Large Language Models under Best-of-N Sampling (NeurIPS 2026 · 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.
