Research map / Architectures and efficiency / Attention and transformers
Densely Connected Attention Propagation for Reading Comprehension
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 |
|---|---|---|---|
| Add the full DecaProp architecture, including dense attention connections (ablation) | Enhanced R-NET · NewsQA dev | F1: 61.2 -> 65.7 | Helped |
| Add DecaCore (ablation) | DecaProp · NewsQA dev | F1: 64.5 -> 65.7 | Helped |
| Add DecaEnc instead of using a 3-layer BiRNN (ablation) | DecaProp · NewsQA dev | F1: 62.0 -> 65.7 | Helped |
| Add cross-hierarchical connections (ablation) | DecaProp · NewsQA dev | F1: 63.1 -> 65.7 | Helped |
| Add gated bi-attention and gated self-attention (ablation) | DecaProp · NewsQA dev | F1: 62.8 -> 65.7 | Helped |
| Use 3 DecaEnc layers rather than 2 (ablation) | DecaProp · NewsQA dev | F1: 63.4 -> 65.7 | Helped |
| Use 3 DecaEnc layers rather than 4 (ablation) | DecaProp · NewsQA dev | F1: 63.3 -> 65.7 | Helped |
| Use factorization-machine compression instead of linear d->1 compression (ablation) | DecaProp BAC · NewsQA dev | F1: 63.0 -> 65.7 | Helped |
| Use factorization-machine compression instead of nonlinear d->d->1 compression (ablation) | DecaProp BAC · NewsQA dev | F1: 60.0 -> 65.7 | Helped |
| Use DecaProp rather than the strongest reported baseline | NewsQA · NewsQA test | F1: +2.6% | Helped |
| Use DecaProp rather than the strongest reported baseline | Quasar-T · Quasar-T test | F1: +6% | Helped |
| Use DecaProp rather than AMANDA | SearchQA original setting · SearchQA original setting | F1: +14.2% | Helped |
Closest papers by meaning
- Hierarchical Attention Flow for Multiple-Choice Reading Comprehension (AAAI 2018 · similarity 0.88)
- Co-Attention Hierarchical Network: Generating Coherent Long Distractors for Reading Comprehension (AAAI 2020 · similarity 0.87)
- Multi-Mention Learning for Reading Comprehension with Neural Cascades (ICLR 2018 · similarity 0.86)
- Bidirectional Attention Flow for Machine Comprehension (ICLR 2017 · similarity 0.86)
- MAttNet: Modular Attention Network for Referring Expression Comprehension (CVPR 2018 · similarity 0.86)
- Virtual Head Attention (NeurIPS 2026 · similarity 0.83)
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.