Research map / Architectures and efficiency / Attention and transformers
Towards Efficient Pre-Trained Language Model via Feature Correlation Distillation
Kun Huang, Xin Guo, Meng Wang
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 Pearson linear correlation instead of KL divergence (ablation) | FCD · MNLI/QQP/RTE | average: 75.1 -> 75.8 | Improved |
| use Pearson linear correlation instead of MSE (ablation) | FCD · MNLI/QQP/RTE | average: 75.5 -> 75.8 | Improved |
| use uniform layer selection instead of top layers (ablation) | FCD · MNLI-m | accuracy: 82.9 -> 83.8 | Improved |
| use uniform layer selection instead of top layers (ablation) | FCD · CoLA | Matthews corr.: 50.7 -> 52.0 | Improved |
| use uniform layer selection instead of bottom layers (ablation) | FCD · MNLI-m | accuracy: 83.3 -> 83.8 | Improved |
| use uniform layer selection instead of bottom layers (ablation) | FCD · CoLA | Matthews corr.: 51.5 -> 52.0 | Improved |
| remove token-level relation loss (ablation) | FCD · SST-2 | score | Hurt |
| remove sample-level relation loss (ablation) | FCD · GLUE (QNLI, MNLI-m, MRPC, SST-2) | score | Hurt |
| distill with FCD instead of KD from BERT-base to 6 layers (ablation) | BERT-small6 · GLUE test average | AVG score: 74.8 -> 79.6 | Improved |
| use FCD instead of MiniLMv2 (BERT teacher) | 6-layer student · GLUE test average | AVG score: 79.2 -> 79.6 | Improved |
| use FCD instead of MiniLMv2 (RoBERTa teacher) | 6-layer student · GLUE test average | AVG score: 81.0 -> 81.9 | Improved |
| use FCD instead of original KD on decoder model | DistilGPT2 · GLUE test average | AVG score: 75.3 -> 77.1 | 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.
- MiniLM6 — MiniLM: Deep Self-Attention Distillation for Task-Agnostic Compression of Pre-Trained Transformers (NeurIPS 2020) · better on 10 of 10 shared results
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
- DistiLLM: Towards Streamlined Distillation for Large Language Models (ICML 2024 · similarity 0.93)
- DA-KD: Difficulty-Aware Knowledge Distillation for Efficient Large Language Models (ICML 2025 · similarity 0.91)
- MiniPLM: Knowledge Distillation for Pre-training Language Models (ICLR 2025 · similarity 0.90)
- DDK: Distilling Domain Knowledge for Efficient Large Language Models (NeurIPS 2024 · similarity 0.89)
- WASD: Wasserstein-based Knowledge Distillation for Large Language Models (NeurIPS 2026 · similarity 0.89)
- TAID: Temporally Adaptive Interpolated Distillation for Efficient Knowledge Transfer in Language Models (ICLR 2025 · similarity 0.89)
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.
