Research map / Probabilistic and scientific modeling / Molecules and proteins
An EM Approach to Non-autoregressive Conditional Sequence Generation
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 |
|---|---|---|---|
| Iteratively update the AR teacher and NAR student with EM training | NAT · WMT14 En-De | BLEU: 19.55 -> 23.27 | Helped |
| Iteratively update the AR teacher and NAR student with EM training | NAT · WMT14 De-En | BLEU: 23.44 -> 26.73 | Helped |
| Iteratively update the AR teacher and NAR student with EM training | NAT · IWSLT14 De-En | BLEU: 22.35 -> 29.38 | Helped |
| Add ODD decoding (ablation) | NAT + EM training · WMT14 En-De | BLEU: 23.27 -> 24.54 | Helped |
| Add ODD decoding (ablation) | NAT + EM training · WMT14 De-En | BLEU: 26.73 -> 27.93 | Helped |
| Add ODD decoding (ablation) | NAT + EM training · IWSLT14 De-En | BLEU: 29.38 -> 30.69 | Helped |
| Use a larger model (ablation) | NAT + EM training + ODD decoding, rescoring 9 · WMT14 En-De | BLEU: 25.75 -> 26.21 | Helped |
| Use a larger model (ablation) | NAT + EM training + ODD decoding, rescoring 9 · WMT14 De-En | BLEU: 29.29 -> 29.81 | Helped |
| Use a larger model (ablation) | NAT + EM training + ODD decoding, rescoring 9 · IWSLT14 De-En | BLEU: 32.66 -> 33.25 | Helped |
| Parameterize the posterior with a unified AR model rather than train NAR directly on the pseudo dataset (ablation) | NAT M-step · WMT14 En-De, iteration t=2 | BLEU: 21.78 -> 22.27 | Helped |
| Parameterize the posterior with a unified AR model rather than train NAR directly on the pseudo dataset (ablation) | NAT M-step · WMT14 En-De, iteration t=3 | BLEU: 22.44 -> 22.85 | Helped |
| Parameterize the posterior with a unified AR model rather than train NAR directly on the pseudo dataset (ablation) | NAT M-step · WMT14 En-De, iteration t=4 | BLEU: 22.98 -> 23.27 | Helped |
Closest papers by meaning
- Fast Structured Decoding for Sequence Models (NeurIPS 2019 · similarity 0.90)
- Improving Non-Autoregressive Translation Models Without Distillation (ICLR 2022 · similarity 0.90)
- SequenceMatch: Imitation Learning for Autoregressive Sequence Modelling with Backtracking (ICLR 2024 · similarity 0.87)
- Tractable Transformers for Flexible Conditional Generation (ICML 2025 · similarity 0.86)
- Beyond Autoregression: Fast LLMs via Self-Distillation Through Time (ICLR 2025 · similarity 0.86)
- Flexible Language Modeling in Continuous Space with Transformer-based Autoregressive Flows (NeurIPS 2025 · similarity 0.85)
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.