Research map / Architectures and efficiency / Pruning and efficient tuning
Adversarial Data Augmentation for Task-Specific Knowledge Distillation of Pre-trained Transformers
Minjia Zhang, Niranjan Uma Naresh, Yuxiong He
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 adversarial embedding augmentation during task-specific distillation | BERT12 → BERT6 · GLUE development set | Avg: 82.4 -> 83.6 | Improved |
| Add task-specific AD2 distillation | MiniLM-L12-H384 · GLUE development set | Avg: 87.1 -> 87.6 | Improved |
| Add deep knowledge distillation (ablation) | AD2, BERT12 → BERT6 · GLUE development set | Avg: 85.9 -> 86.2 | Improved |
| Add knowledge distillation (ablation) | AD2, BERT12 → BERT6 · GLUE development set | Avg: 85.2 -> 85.9 | Improved |
| Add adversarial data augmentation (ablation) | AD2, BERT12 → BERT6 · GLUE development set | Avg: 84.3 -> 85.2 | Improved |
| Initialize the student from bottom rather than top layers (ablation) | 3-layer student initialized from DistilBERT · GLUE development set | Avg: 75.6 -> 80.6 | Improved |
| Add AD2 to bottom-layer initialization (ablation) | 3-layer student initialized from DistilBERT · GLUE development set | Avg: 80.6 -> 82.3 | Improved |
| Replace BERT12 with BERT24 as teacher while using KD without AD2 (ablation) | BERT6 student · GLUE development set | Avg: 84.2 -> 84.0 | No clear effect |
| Use BERT24 as teacher and add AD2 (ablation) | BERT6 student · GLUE development set | Avg: 84.0 -> 85.3 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- MiniLM — MiniLM: Deep Self-Attention Distillation for Task-Agnostic Compression of Pre-Trained Transformers (NeurIPS 2020) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Adversarial Moment-Matching Distillation of Large Language Models (NeurIPS 2024 · similarity 0.89)
- Impartial Adversarial Distillation: Addressing Biased Data-Free Knowledge Distillation via Adaptive Constrained Optimization (AAAI 2024 · similarity 0.88)
- MiniPLM: Knowledge Distillation for Pre-training Language Models (ICLR 2025 · similarity 0.87)
- DistiLLM: Towards Streamlined Distillation for Large Language Models (ICML 2024 · similarity 0.86)
- Hybrid Data-Free Knowledge Distillation (AAAI 2025 · similarity 0.86)
- DA-KD: Difficulty-Aware Knowledge Distillation for Efficient Large Language Models (ICML 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.
