Research map / Architectures and efficiency / Attention and transformers
GRET: Global Representation Enhanced Transformer
Rongxiang Weng, Haoran Wei, Shujian Huang, Heng Yu, Lidong Bing, Weihua Luo, Jiajun Chen
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 |
|---|---|---|---|
| Add global representation modeling and decoder fusion | Transformer · WMT17 ZH→EN newstest2017 | BLEU: 24.31 -> 25.53 | Improved |
| Add global representation modeling and decoder fusion | Transformer · WMT14 EN→DE newstest2014 | BLEU: 27.20 -> 28.46; DeepRepre 28.78 | Mixed |
| Add global representation modeling and decoder fusion | Transformer · WMT14 DE→EN newstest2014 | BLEU: 32.34 -> 33.79 | Improved |
| Add global representation modeling and decoder fusion | Transformer · WMT16 RO→EN newstest2016 | BLEU: 32.17 -> 33.06; 0.89 BLEU | Improved |
| Add global representation modeling and decoder fusion | Transformer · LCSTS test | ROUGE-1: 43.14 -> 44.77; +1.63 | Improved |
| Add global representation modeling and decoder fusion | Transformer · LCSTS test | ROUGE-2: 29.26 -> 30.96; +1.70 | Improved |
| Add global representation modeling and decoder fusion | Transformer · LCSTS test | ROUGE-L: 39.72 -> 41.21; +1.49 | Improved |
| Add capsule feature extraction (ablation) | GRET · WMT14 EN→DE newstest2014 | BLEU: 27.76 -> 28.46 | Improved |
| Add layer-wise aggregation (ablation) | GRET · WMT14 EN→DE newstest2014 | BLEU: 27.81 -> 28.46 | Improved |
| Add context gating (ablation) | GRET · WMT14 EN→DE newstest2014 | BLEU: 28.23 -> 28.46 | Improved |
| Use the Big model setting | GRET · WMT14 EN→DE newstest2014 | BLEU: 28.47 -> 29.33; 0.86 BLEU | Improved |
| Increase dynamic-routing iterations beyond the first three (ablation) | GRET · WMT14 EN→DE newstest2014 | BLEU: 4 and 5 | No clear effect |
Closest papers by meaning
- Deliberation Networks: Sequence Generation Beyond One-Pass Decoding (NeurIPS 2017 · similarity 0.85)
- Transformers need glasses! Information over-squashing in language tasks (NeurIPS 2024 · similarity 0.84)
- Trainable Transformer in Transformer (ICML 2024 · similarity 0.84)
- Bigram Subnetworks: Mapping to Next Tokens in Transformer Language Models (NeurIPS 2025 · similarity 0.84)
- Memory-Augmenting Decoder-Only Language Models through Encoders (Student Abstract) (AAAI 2024 · similarity 0.84)
- Transformer Block Coupling and its Correlation with Generalization in LLMs (ICLR 2025 · similarity 0.84)
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.
