Research map / Architectures and efficiency / Attention and transformers
Adversarial Self-Attention for Language Understanding
Hongqiu Wu, Ruixue Ding, Hai Zhao, Pengjun Xie, Fei Huang, Min Zhang
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 |
|---|---|---|---|
| Add ASA fine-tuning to BERT | BERT · SST-2 | accuracy: 93.2 -> 94.1 | Improved |
| Add ASA fine-tuning to BERT | BERT · MNLI | accuracy: 84.1 -> 85.0 | Improved |
| Add ASA fine-tuning to BERT | BERT · QNLI | accuracy: 90.4 -> 91.4 | Improved |
| Add ASA fine-tuning to RoBERTa | RoBERTa · QQP | F1: 72.2 -> 73.7 | Improved |
| Add ASA fine-tuning to BERT | BERT · STS-B | Spearman correlation: 84.7 to 86.5 | Improved |
| Add ASA fine-tuning to BERT | BERT · WNUT-17 | F1: 47.8 -> 49.8 | Improved |
| Add ASA fine-tuning to BERT | BERT · DREAM | accuracy: 62.9 to 64.3 | Improved |
| Add ASA fine-tuning to BERT | BERT · ANLI | accuracy: 48.0 -> 50.4 | Improved |
| Add ASA fine-tuning to BERT | BERT · PAWS-QQP | accuracy: 81.7 -> 87.7 | Improved |
| Add ASA fine-tuning to BERT | BERT · HellaSWAG | accuracy: 39.7 -> 40.8 | Improved |
| Replace ASA with random Bernoulli attention masking (ablation) | ASA · PAWS-QQP | accuracy: 87.7 vs 86.1 | Hurt |
| Replace ASA with scheduled Bernoulli masking (ablation) | ASA · PAWS-QQP | accuracy: 87.7 vs 85.4 | Hurt |
Closest papers by meaning
- The Closeness of In-Context Learning and Weight Shifting for Softmax Regression (NeurIPS 2024 · similarity 0.86)
- Theory, Analysis, and Best Practices for Sigmoid Self-Attention (ICLR 2025 · similarity 0.86)
- Selective Attention: Enhancing Transformer through Principled Context Control (NeurIPS 2024 · similarity 0.86)
- From Attention to Activation: Unraveling the Enigmas of Large Language Models (ICLR 2025 · similarity 0.85)
- Understanding Differential Transformer Unchains Pretrained Self-Attentions (NeurIPS 2025 · similarity 0.85)
- KV Shifting Attention Enhances Language Modeling (ICML 2025 · similarity 0.85)
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.
