Research map / Adversarial robustness and security / Certified and empirical robustness
A Robust Adversarial Training Approach to Machine Reading Comprehension
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 |
|---|---|---|---|
| adversarial training with model-generated perturbation-based misleading texts | ERNIE2.0 · average of adversarial test sets | F1: 75.5 -> 83.9 | Helped |
| adversarial training with generated misleading texts | BERTbase · average of adversarial test sets | F1: 59.0 -> 73.0 | Helped |
| adversarial training with generated misleading texts | QANet · average of adversarial test sets | F1: 64.9 -> 68.0 | Helped |
| adversarial training with generated misleading texts | ERNIE2.0 · SQuAD dev | F1: 92.5 -> 92.5 | No clear effect |
| add the approach on top of AddSentDiverse training data | ERNIE2.0+AddSentDiverse · average of adversarial test sets | F1: 86.0 -> 87.4 | Helped |
| rule-based AddSentDiverse augmentation vs ours on unseen adversarial types (AAE, AAC, ANA) | MRC models · AAE, AAC, ANA | F1: +1.7 (ASD) vs +4.7 (ours) avg F1 | Mixed |
| add answer-type misleading loss La (ablation) | BERTbase+Ours · average adversarial F1 | F1: 64.1 -> 73.0 | Helped |
| add context-type misleading loss Lc (ablation) | BERTbase+Ours · average adversarial F1 | F1: 70.9 -> 73.0 | Helped |
| add similarity regularizer Rs (ablation) | BERTbase+Ours · average adversarial F1 | F1: 69.7 -> 73.0 | Helped |
| increase adversarial training iterations beyond 4-5 (ablation) | BERTbase+Ours · average adversarial F1 | F1 | No clear effect |
Closest papers by meaning
- Multi-Task Learning with Generative Adversarial Training for Multi-Passage Machine Reading Comprehension (AAAI 2020 · similarity 0.88)
- Perturbation-Invariant Adversarial Training for Neural Ranking Models: Improving the Effectiveness-Robustness Trade-Off (AAAI 2024 · similarity 0.84)
- One Prompt Word is Enough to Boost Adversarial Robustness for Pre-trained Vision-Language Models (CVPR 2024 · similarity 0.84)
- Towards Robust Knowledge Unlearning: An Adversarial Framework for Assessing and Improving Unlearning Robustness in Large Language Models (AAAI 2025 · similarity 0.84)
- DAT: Improving Adversarial Robustness via Generative Amplitude Mix-up in Frequency Domain (NeurIPS 2024 · similarity 0.83)
- A Closer Look at Curriculum Adversarial Training: From an Online Perspective (AAAI 2024 · similarity 0.83)
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.