Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
Tuning Language Models as Training Data Generators for Augmentation-Enhanced Few-Shot Learning
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 |
|---|---|---|---|
| Use FewGen augmentation rather than the strongest no-augmentation method, LM-BFF (Auto) | Few-shot classification on GLUE · GLUE seven-task average | average score: 65.8 -> 72.8 | Helped |
| Use FewGen augmentation rather than GPT3Mix | Few-shot classification on GLUE · GLUE seven-task average | average score: 69.2 -> 72.8 | Helped |
| Use FewGen augmentation rather than back translation | Few-shot classification on GLUE · GLUE seven-task average | average score: 60.6 -> 72.8 | Helped |
| Add meta-weighted generator tuning instead of standard maximum likelihood generator tuning (ablation) | FewGen generator · GLUE classification tasks | classification performance: 74.9 -> 75.7 (MNLI-m) | Helped |
| Add meta-weighted generator tuning instead of directly combining generative and discriminative losses (ablation) | FewGen generator · GLUE classification tasks | classification performance: 74.6 -> 75.7 (MNLI-m) | Helped |
| Add label smoothing during synthetic-data classifier fine-tuning (ablation) | FewGen classifier · MNLI | accuracy: 75.0 -> 75.7 | No clear effect |
| Add temporal ensembling during synthetic-data classifier fine-tuning (ablation) | FewGen classifier · MNLI | accuracy: 72.2 -> 75.7 | Helped |
| Use the two-step procedure: train on few-shot data, then fine-tune on generated data, instead of joint fine-tuning (ablation) | FewGen classifier · MNLI | accuracy: 68.9 -> 75.7 | Helped |
| Use meta-weighted generator objective rather than standard language modeling objective (ablation) | FewGen generator · MNLI | generated data accuracy: 69.4 -> 72.3 | Helped |
| Use meta-weighted generator objective rather than directly combining generative and discriminative losses (ablation) | FewGen generator · MNLI | generated data accuracy: 70.2 -> 72.3 | Helped |
Closest papers by meaning
- Optimization Inspired Few-Shot Adaptation for Large Language Models (NeurIPS 2025 · similarity 0.90)
- Many-Shot In-Context Learning (NeurIPS 2024 · similarity 0.89)
- The Surprising Effectiveness of Test-Time Training for Few-Shot Learning (ICML 2025 · similarity 0.88)
- Rethinking Few-Shot Adaptation of Vision-Language Models in Two Stages (CVPR 2025 · similarity 0.88)
- Drag-and-Drop LLMs: Zero-Shot Prompt-to-Weights (NeurIPS 2025 · similarity 0.88)
- Making Large Vision Language Models to Be Good Few-Shot Learners (AAAI 2025 · similarity 0.87)
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.