Research map / Architectures and efficiency / Pruning and efficient tuning
Acquiring Knowledge from Pre-Trained Model to 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 |
|---|---|---|---|
| Use APT with BERT encoder and GPT decoder instead of the Transformer baseline | Transformer NMT · WMT14 EN→DE test | BLEU: 27.31 -> 29.23 | Helped |
| Use APT with BERT encoder and GPT decoder instead of the Transformer baseline | Transformer NMT · WMT14 DE→EN test | BLEU: 32.51 -> 34.84 | Helped |
| Use APT with BERT encoder and BERT decoder instead of the Transformer baseline | Transformer NMT · WMT17 ZH→EN test | BLEU: 24.47 -> 26.46 | Helped |
| Add knowledge distillation (ablation) | APT, EN→DE · WMT14 EN→DE test | BLEU: 28.68 -> 29.23 | Helped |
| Add contextual gating (ablation) | APT dynamic fusion, EN→DE · WMT14 EN→DE test | BLEU: 28.44 -> 28.77 | Helped |
| Add layer-aware attention (ablation) | APT dynamic fusion, EN→DE · WMT14 EN→DE test | BLEU: 28.39 -> 28.77 | Helped |
| Add word-level distillation (ablation) | APT with dynamic fusion, EN→DE · WMT14 EN→DE test | BLEU: 28.76 -> 29.23 | Helped |
| Add sentence-level distillation (ablation) | APT with dynamic fusion, EN→DE · WMT14 EN→DE test | BLEU: 28.87 -> 29.23 | Helped |
| Use dynamic fusion rather than knowledge distillation on the encoder (ablation) | APT encoder integration · WMT14 EN→DE test | BLEU: 28.21 -> 28.77 | Helped |
| Use knowledge distillation rather than dynamic fusion on the decoder (ablation) | APT decoder integration · WMT14 EN→DE test | BLEU: 27.41 -> 28.68 | Helped |
| Combine dynamic fusion with knowledge distillation on the encoder (ablation) | APT encoder integration · WMT14 EN→DE test | BLEU: 28.77 -> 28.69 | Hurt |
| Combine dynamic fusion with knowledge distillation on the decoder (ablation) | APT decoder integration · WMT14 EN→DE test | BLEU: 28.68 -> 27.71 | Hurt |
Closest papers by meaning
- Towards Making the Most of BERT in Neural Machine Translation (AAAI 2020 · similarity 0.93)
- Deep Fusing Pre-trained Models into Neural Machine Translation (AAAI 2022 · similarity 0.91)
- Translating with Bilingual Topic Knowledge for Neural Machine Translation (AAAI 2019 · similarity 0.89)
- Multilingual Neural Machine Translation with Knowledge Distillation (ICLR 2019 · similarity 0.89)
- Seeking Neural Nuggets: Knowledge Transfer in Large Language Models from a Parametric Perspective (ICLR 2024 · similarity 0.88)
- A Paradigm Shift in Machine Translation: Boosting Translation Performance of Large Language Models (ICLR 2024 · similarity 0.87)
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.