Research map / Large language models / Language, safety and interpretability
Few-Shot Learning Evaluation in Natural Language Understanding
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 prompt-based rather than classic fine-tuning | RoBERTa Large · SST-2, 10-shot | S1: 38.4 { (21.7)} -> 88.8 { (3.9)} | Helped |
| Use prompt-based rather than classic fine-tuning | RoBERTa Large · MNLI, 10-shot | S1: 34.3 { (2.8)} -> 57.7 { (3.6)} | Helped |
| Use GPT-3 in-context learning rather than few-shot training baselines | SST-2 · SST-2, 10-shot | S1: 85.9 { (3.7)} | Helped |
| Use GPT-3 in-context learning | MNLI · MNLI, 10-shot | S1: 33.5 { (0.7)} | Hurt |
| Increase prompt fine-tuning examples from 10 to 30 | RoBERTa Large · SST-2 | S1: 88.8 { (3.9)} -> 90.2 { (1.8)} | Helped |
| Increase prompt fine-tuning examples from 10 to 30 | RoBERTa Large · MNLI | S1: 57.7 { (3.6)} -> 61.6 { (3.5)} | Helped |
| Increase classic fine-tuning examples in the few-shot setting | PLMs · few-shot tasks | S1 | No clear effect |
| Use prompt-based fine-tuning rather than classic fine-tuning in the fully supervised setting | classification models · SST-2 and MNLI, All | S1 | No clear effect |
| Increase GPT-3 in-context demonstrations from 10 to 20 | GPT-3 · SST-2 | S1: 85.9 { (3.7)} -> 92.0 { (0.7)} | Helped |
| Increase model size | PLMs with classic fine-tuning · few-shot tasks | S1 | Mixed |
| Increase model size | PLMs with prompt-based fine-tuning · SST-2 and MNLI | S1 | Helped |
| Use GPT-3 in-context learning on structured NER and MRC tasks | GPT-3 · CoNLL03, WikiANN, and SQuAD v2 | S1: 4.7; 1.8; 2.5 | Hurt |
Closest papers by meaning
- The Surprising Effectiveness of Test-Time Training for Few-Shot Learning (ICML 2025 · similarity 0.89)
- Evaluating the Zero-shot Robustness of Instruction-tuned Language Models (ICLR 2024 · similarity 0.87)
- Many-Shot In-Context Learning (NeurIPS 2024 · similarity 0.87)
- Improving Natural Language Understanding for LLMs via Large-Scale Instruction Synthesis (AAAI 2025 · similarity 0.87)
- Descriptor and Word Soups: Overcoming the Parameter Efficiency Accuracy Tradeoff for Out-of-Distribution Few-shot Learning (CVPR 2024 · similarity 0.86)
- Task Contamination: Language Models May Not Be Few-Shot Anymore (AAAI 2024 · 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.