Research map / Label-efficient and robust learning / Recognition and long-tail
Multi-Mask Label Mapping for Prompt-Based 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 MMLM weighted voting instead of KPT label mapping | RoBERTa-large prompt classification · AG’s News, K=1 and K=20 | accuracy: 82.8 -> 83.0; 87.5 -> 88.6 | Helped |
| Use MMLM weighted voting instead of KPT label mapping | RoBERTa-large prompt classification · IMDB, K=1 and K=20 | accuracy: 92.0 -> 92.6; 94.1 -> 94.5 | Helped |
| Use MMLM weighted voting instead of KPT label mapping | RoBERTa-large prompt classification · Amazon, K=1 and K=20 | accuracy: 92.1 -> 92.4; 94.4 -> 95.4 | Helped |
| Use MMLM weighted voting instead of KPT label mapping | RoBERTa-large prompt classification · DBPedia, K=1 and K=20 | accuracy: 94.9 -> 95.1; 96.9 -> 97.1 | Helped |
| Use MMLM weighted voting instead of KPT label mapping | RoBERTa-large prompt classification · Yahoo, K=1 and K=20 | accuracy: 53.9 -> 54.7; 60.7 -> 61.3 | Helped |
| Add dropout layers to the multi-mask scheme (ablation) | MMLM on AG’s News · AG’s News, K=1 through K=20 | accuracy: 82.8 -> 83.0; 85.4 -> 85.6; 87.0 -> 87.1; 88.2 -> 88.6 | Helped |
| Add dropout layers and embedding disturbance (ablation) | MMLM on AG’s News · AG’s News, K=1 through K=20 | accuracy: 82.7 -> 83.0; 85.3 -> 85.6; 86.9 -> 87.1; 88.0 -> 88.6 | Helped |
| Use multiple classifiers rather than one classifier (ablation) | MMLM on AG’s News · AG’s News, K=1 through K=20 | accuracy: 82.4 -> 83.0; 84.9 -> 85.6; 86.4 -> 87.1; 87.2 -> 88.6 | Helped |
| Add keyword augmentation and multiple classifiers (ablation) | Vanilla label mapping on AG’s News · AG’s News, K=1 through K=20 | accuracy: 78.4 -> 83.0; 83.1 -> 85.6; 85.1 -> 87.1; 86.4 -> 88.6 | Helped |
| Use separate multi-mask classifiers instead of packing all keywords into one classifier (ablation) | Prompt-based cloze augmentation on AG’s News · AG’s News, K=1 through K=20 | accuracy: 81.3 -> 83.0; 84.9 -> 85.6; 86.7 -> 87.1; 87.9 -> 88.6 | Helped |
| Use MMLM weighted voting instead of KPT label mapping | RoBERTa-large with NullPrompt · AG’s News, K=1 through K=20 | accuracy: 76.4 -> 77.5; 82.6 -> 83.2; 85.6 -> 86.2; 86.7 -> 87.6 | Helped |
Closest papers by meaning
- Multi-Label Few-Shot Image Classification via Pairwise Feature Augmentation and Flexible Prompt Learning (AAAI 2025 · similarity 0.86)
- Labels Need Prompts Too: Mask Matching for Natural Language Understanding Tasks (AAAI 2024 · similarity 0.86)
- LAMM: Label Alignment for Multi-Modal Prompt Learning (AAAI 2024 · similarity 0.85)
- SEC-Prompt:SEmantic Complementary Prompting for Few-Shot Class-Incremental Learning (CVPR 2025 · similarity 0.85)
- Correlative and Discriminative Label Grouping for Multi-Label Visual Prompt Tuning (CVPR 2025 · similarity 0.85)
- Learning from True-False Labels via Multi-modal Prompt Retrieving (ICML 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.