Research map / Architectures and efficiency / Attention and transformers
Neural Simile Recognition with Cyclic Multitask Learning and Local Attention
Jiali Zeng, Linfeng Song, Jinsong Su, Jun Xie, Wei Song, Jiebo Luo
Read the paperCheck your idea against 137,217 papers
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 |
|---|---|---|---|
| replace global attention with local attention around the comparator (ablation) | simile sentence classifier · Chinese simile classification | F1-score: 82.84 -> 83.79 | Improved |
| connect subtasks cyclically instead of using conventional shared-task MTL (ablation) | simile classification · Chinese simile sentence classification | F1-score: 86.53 -> 87.04 | Improved |
| connect subtasks cyclically instead of using conventional shared-task MTL (ablation) | simile component extraction · Chinese simile component extraction | F1-score: 62.68 -> 68.05 | Improved |
| increase cyclic execution from K=0 to K=1 (ablation) | MTL-Cyc · Chinese simile classification, validation | F1-score: 86.09 -> 86.62 | Improved |
| increase cyclic execution from K=0 to K=1 (ablation) | MTL-Cyc · Chinese simile component extraction, validation | F1-score: 63.15 -> 73.33 | Improved |
| increase cyclic execution from K=1 to K=2 (ablation) | MTL-Cyc · Chinese simile classification and component extraction, validation | F1-score: 86.62 -> 86.65; 73.33 -> 73.40 | No clear effect |
| increase cyclic execution beyond K=2 (ablation) | MTL-Cyc · Chinese simile classification and component extraction, validation | F1-score: 86.65 -> 85.89; 73.40 -> 72.83 | Hurt |
| add sentence reconstruction as an auxiliary task (ablation) | MTL simile component extractor · Chinese simile component extraction | F1-score: 59.89 -> 63.29 | Improved |
| order the pipeline as classification → extraction rather than extraction → classification (ablation) | MTL-Pip · Chinese simile sentence classification | F1-score: 85.20 -> 85.51 | Improved |
| order the pipeline as classification → extraction rather than extraction → classification (ablation) | MTL-Pip · Chinese simile component extraction | F1-score: 61.84 -> 66.45 | Improved |
| use cyclic connections rather than a classification → extraction → reconstruction pipeline (ablation) | simile classifier · Chinese simile sentence classification | F1-score: 86.81 -> 87.04 | Improved |
| use cyclic connections rather than a classification → extraction → reconstruction pipeline (ablation) | simile component extractor · Chinese simile component extraction | F1-score: 67.63 -> 68.05 | Improved |
Closest papers by meaning
- Writing Polishment with Simile: Task, Dataset and A Neural Approach (AAAI 2021 · similarity 0.84)
- MAPS-KB: A Million-Scale Probabilistic Simile Knowledge Base (AAAI 2023 · similarity 0.83)
- MmAP: Multi-Modal Alignment Prompt for Cross-Domain Multi-Task Learning (AAAI 2024 · similarity 0.81)
- Octopus: A Multi-modal LLM with Parallel Recognition and Sequential Understanding (NeurIPS 2024 · similarity 0.80)
- Distribution Matching for Multi-Task Learning of Classification Tasks: A Large-Scale Study on Faces & Beyond (AAAI 2024 · similarity 0.80)
- Training Dynamics of Transformers to Recognize Word Co-occurrence via Gradient Flow Analysis (NeurIPS 2024 · similarity 0.80)
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.
