Research map / Adversarial robustness and security / Attacks on text and graphs
The King Is Naked: On the Notion of Robustness for Natural Language Processing
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 |
|---|---|---|---|
| Add shallow-negation augmentation | LSTM · Shallow Negation | accuracy: 0.4771 ± 0.0143 -> 0.6387 ± 0.0387 | Helped |
| Add shallow-negation augmentation | Self-attention · Shallow Negation | accuracy: 0.4790 ± 0.0059 -> 0.5954 ± 0.0027 | Helped |
| Add mixed-sentiment augmentation | LSTM · Mixed Sentiment | accuracy: 0.5110 ± 0.0251 -> 0.6357 ± 0.0317 | Helped |
| Add mixed-sentiment augmentation | Self-attention · Mixed Sentiment | accuracy: 0.5487 ± 0.0099 -> 0.5617 ± 0.0048 | No clear effect |
| Add sarcasm augmentation | Self-attention · Sarcasm | accuracy: 0.5240 ± 0.0132 -> 0.6236 ± 0.0218 | Helped |
| Add sarcasm augmentation | LSTM · Sarcasm | accuracy: 0.5578 ± 0.0128 -> 0.4807 ± 0.0197 | Hurt |
| Increase augmentation factor from 500 to 750 (ablation) | LSTM and self-attention models · Shallow Negation, Mixed Sentiment, and Sarcasm | accuracy | Mixed |
| Increase augmentation factor to 1K or greater (ablation) | All tested architectures · Semantic robustness tasks | accuracy | Hurt |
| Replace vanilla training with shallow-negation semantic augmentation | Neural networks · Shallow Negation test set | accuracy: 0.4916 ± 0.0074 -> 0.5491 ± 0.0124 | Helped |
| Replace vanilla training with shallow-negation semantic augmentation | Neural networks · SST-2 test set | accuracy: 0.9036 ± 0.0019 -> 0.8838 ± 0.0049 | Hurt |
| Use IBP training at ε=0.001 instead of vanilla training (ablation) | FC · Shallow Negation (Our Benchmark) | accuracy: 0.4034 ± 0.0214 -> 0.3852 ± 0.0071 | Hurt |
| Use IBP training at ε=0.01 instead of vanilla training (ablation) | FC · Shallow Negation (Our Benchmark) | accuracy: 0.4034 ± 0.0214 -> 0.4249 ± 0.0260 | Helped |
Closest papers by meaning
- Remarkable Robustness of LLMs: Stages of Inference? (NeurIPS 2025 · similarity 0.89)
- Evaluating the Zero-shot Robustness of Instruction-tuned Language Models (ICLR 2024 · similarity 0.87)
- Interpretable but Fragile? Robustness of Concept Bottlenecks under Geometric-Semantic Perturbations (NeurIPS 2026 · similarity 0.87)
- Towards Robust Knowledge Unlearning: An Adversarial Framework for Assessing and Improving Unlearning Robustness in Large Language Models (AAAI 2025 · similarity 0.87)
- Brittlebench: Quantifying LLM robustness via prompt sensitivity (NeurIPS 2026 · similarity 0.86)
- On the Robustness of Verbal Confidence of LLMs in Adversarial Attacks (NeurIPS 2025 · similarity 0.86)
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.