Research map / Architectures and efficiency / Pruning and efficient tuning
Deep Fusing Pre-trained Models into Neural Machine Translation
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 |
|---|---|---|---|
| deep-fuse mBERT using layer-wise coordination and partitioned multi-task training | Transformer NMT · WMT14 En→De | BLEU: 29.82 -> 31.59 | Helped |
| deep-fuse mBERT using layer-wise coordination and partitioned multi-task training | Transformer NMT · WMT14 En→Fr | BLEU: 42.93 -> 44.21 | Helped |
| deep-fuse mBERT in the Mask-Predict model | NAT · WMT14 En→De | BLEU: +2.14 | Helped |
| add back-translation data | Deep-fused NMT · WMT14 En→De | BLEU: 31.59 -> 32.56 | Helped |
| add back-translation data | Deep-fused NMT · WMT16 Ro→En | BLEU: 36.71 -> 39.48 | Helped |
| add target words prediction (ablation) | Deep-fused NMT · WMT14 En→De | BLEU: 31.21 -> 31.59 | Helped |
| add target sentence prediction (ablation) | Deep-fused NMT · WMT14 En→De | BLEU: 31.17 -> 31.59 | Helped |
| add both target words prediction and target sentence prediction (ablation) | Deep-fused NMT · WMT14 En→De | BLEU: 30.94 -> 31.59 | Helped |
| retain pre-training (ablation) | Deep-fused NMT · WMT14 En→De | BLEU: 30.17 -> 31.59 | Helped |
| use the default multilingual BERT rather than monolingual BERT (ablation) | Deep-fused NMT · WMT14 En→De | BLEU: 30.99 -> 31.59 | Helped |
| restrict PTM optimization to the designated intermediate tasks and MLM, rather than all tasks (ablation) | Deep-fused NMT · WMT14 En→De | BLEU: 30.62 -> 31.59 | Helped |
| fine-tune rather than freeze PTM parameters (ablation) | Deep-fused NMT · WMT14 En→De | BLEU: 30.98 -> 31.59 | Helped |
Closest papers by meaning
- Acquiring Knowledge from Pre-Trained Model to Neural Machine Translation (AAAI 2020 · similarity 0.91)
- Towards Making the Most of BERT in Neural Machine Translation (AAAI 2020 · similarity 0.90)
- Layer-Wise Coordination between Encoder and Decoder for Neural Machine Translation (NeurIPS 2018 · similarity 0.90)
- Neural Machine Translation with Joint Representation (AAAI 2020 · similarity 0.90)
- Regularizing Neural Machine Translation by Target-Bidirectional Agreement (AAAI 2019 · similarity 0.89)
- A Paradigm Shift in Machine Translation: Boosting Translation Performance of Large Language Models (ICLR 2024 · similarity 0.88)
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.