Research map / Label-efficient and robust learning / Knowledge distillation
Reinforced Multi-Teacher Selection for Knowledge Distillation
Fei Yuan, Linjun Shou, Jian Pei, Wutao Lin, Ming Gong, Yan Fu, Daxin Jiang
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 |
|---|---|---|---|
| Select teachers dynamically per instance using reward3, including development-set accuracy | RL-KD with BERT3 · GLUE task average | accuracy: 77.5 -> 79.6 | Improved |
| Select teachers dynamically per instance using reward3, including development-set accuracy | RL-KD with BERT6 · GLUE task average | accuracy: 82.4 -> 83.8 | Improved |
| Add negative distillation loss to the reward (ablation) | RL-KD with BERT3 · GLUE task average | accuracy: 79.2 -> 79.4 | Improved |
| Add negative distillation loss to the reward (ablation) | RL-KD with BERT6 · GLUE task average | accuracy: 83.6 -> 83.5 | Hurt |
| Add development-set accuracy to the reward (ablation) | RL-KD with BERT3 · GLUE task average | accuracy: 79.4 -> 79.6 | Improved |
| Add development-set accuracy to the reward (ablation) | RL-KD with BERT6 · GLUE task average | accuracy: 83.5 -> 83.8 | Improved |
| Use reinforced teacher selection rather than W-Ensemble | Student distillation · RTE | accuracy: 57.32 -> 63.60 | Improved |
| Use reinforced teacher selection rather than LR-Ensemble | Student distillation · MRPC | accuracy: 73.42 -> 75.52 | Improved |
Closest papers by meaning
- DistiLLM: Towards Streamlined Distillation for Large Language Models (ICML 2024 · similarity 0.90)
- TAID: Temporally Adaptive Interpolated Distillation for Efficient Knowledge Transfer in Language Models (ICLR 2025 · similarity 0.89)
- DA-KD: Difficulty-Aware Knowledge Distillation for Efficient Large Language Models (ICML 2025 · similarity 0.88)
- On Teacher Hacking in Language Model Distillation (ICML 2025 · similarity 0.88)
- Distilling What Matters: Confidence-Aware Selective Distillation for Large Language Models (NeurIPS 2026 · similarity 0.88)
- DDK: Distilling Domain Knowledge for Efficient Large Language Models (NeurIPS 2024 · similarity 0.88)
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.
