Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
Few-shot Text Classification with Distributional Signatures
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 distributional-signature attention with a ridge regressor instead of the strongest listed baseline | Few-shot text and relation classification · Six benchmarks, 5-way 1-shot | accuracy: 60.1 vs. 52.6 average | Helped |
| Use distributional-signature attention with a ridge regressor instead of the strongest listed baseline | Few-shot text and relation classification · Six benchmarks, 5-way 5-shot | accuracy: 78.0 vs. 74.1 average | Helped |
| Add the class-specific statistic t(x) (ablation) | Distributional Signatures · Six-dataset average, 5-way 1-shot | accuracy: 58.1 -> 60.1 | Helped |
| Add the general statistic s(x) (ablation) | Distributional Signatures · Six-dataset average, 5-way 1-shot | accuracy: 54.2 -> 60.1 | Helped |
| Fuse signatures across the input with a BiLSTM instead of scoring each word independently (ablation) | Distributional Signatures · Six-dataset average, 5-way 1-shot | accuracy: 58.3 -> 60.1 | Helped |
| Restrict attention-generator input to distributional signatures rather than also feeding lexical embeddings (ablation) | Distributional Signatures · Six-dataset average, 5-way 1-shot | accuracy: 54.8 -> 60.1 | Helped |
| Use BERT contextualized representations instead of fastText embeddings | Distributional Signatures · HuffPost and FewRel | accuracy: HuffPost: 43.0 -> 42.12 (1-shot); 63.5 -> 62.97 (5-shot); FewRel: 67.1 -> 70.08 (1-shot); 83.5 -> 88.07 (5-shot) | Mixed |
| Finetune word embeddings during meta-training instead of fixing them | Distributional Signatures · Reuters and Amazon, 5-way 5-shot | accuracy: Reuters: 96.00 -> 95.98; Amazon: 81.16 -> 78.94 | Hurt |
Closest papers by meaning
- Adaptive Cross-Modal Few-shot Learning (NeurIPS 2019 · similarity 0.88)
- Unraveling Meta-Learning: Understanding Feature Representations for Few-Shot Tasks (ICML 2020 · similarity 0.88)
- Liberating Seen Classes: Boosting Few-Shot and Zero-Shot Text Classification via Anchor Generation and Classification Reframing (AAAI 2024 · similarity 0.87)
- Text Augmented Correlation Transformer For Few-shot Classification & Segmentation (CVPR 2025 · similarity 0.86)
- Few-Shot Recognition via Stage-Wise Retrieval-Augmented Finetuning (CVPR 2025 · similarity 0.85)
- Making Large Vision Language Models to Be Good Few-Shot Learners (AAAI 2025 · similarity 0.85)
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.