Research map / Learning theory and algorithms / Prediction and decision losses
Noisy-Channel Minimum Bayes Risk Decoding
Yusuke Sakai, Hidetaka Kamigaito, Taro Watanabe
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 |
|---|---|---|---|
| increase hypothesis-to-reference likelihood weight alpha (ablation) | MBR decoding with BLEU or chrF utility · WMT22/23 translation, 12 scenarios | z-score of metric | Hurt |
| increase hypothesis-to-reference likelihood weight alpha (ablation) | MBR decoding with BERTScore utility · WMT22/23 translation, 12 scenarios | z-score of metric | Improved |
| increase hypothesis prior weight beta above 1 (ablation) | MBR decoding with BLEU or chrF utility · WMT22/23 translation | z-score of metric | Improved |
| vary hypothesis prior weight beta (ablation) | MBR decoding with COMET or BERTScore utility · WMT22/23 translation | z-score of metric | No clear effect |
| raise reference-to-hypothesis likelihood weight gamma above 1 (ablation) | MBR decoding, all metrics except BERTScore · WMT22/23 translation | z-score of metric | Hurt |
| lower gamma to 0.25 or 0.5 (ablation) | MBR decoding with BLEU or chrF utility · WMT22/23 translation | z-score of metric | Mixed |
| increase reference prior weight delta (ablation) | MBR decoding with BLEU or BERTScore utility · WMT22/23 translation | z-score of metric | Improved |
| vary reference prior weight delta (ablation) | MBR decoding with chrF or COMET utility · WMT22/23 translation | z-score of metric | Mixed |
| raise alpha and delta together (ablation) | MBR decoding with BERTScore · summarization, captioning, translation | z-score of BERTScore | Improved |
| set alpha above 1.5 (ablation) | MBR decoding with BERTScore · summarization, captioning, translation | z-score of BERTScore | Mixed |
| reweight channels using best weights from WMT22 En-De on WMT23 En-De | standard MBR with chrF · WMT23 En-De | chrF: 50.05 -> 50.18 | Improved |
Closest papers by meaning
- Model-Based Minimum Bayes Risk Decoding for Text Generation (ICML 2024 · similarity 0.93)
- Uncertainty-Aware Decoding with Minimum Bayes Risk (ICLR 2025 · similarity 0.91)
- Efficient Minimum Bayes Risk Decoding using Low-Rank Matrix Completion Algorithms (NeurIPS 2024 · similarity 0.88)
- MBR and QE Finetuning: Training-time Distillation of the Best and Most Expensive Decoding Methods (ICLR 2024 · similarity 0.86)
- Better Instruction-Following Through Minimum Bayes Risk (ICLR 2025 · similarity 0.86)
- A Theoretical Perspective for Speculative Decoding Algorithm (NeurIPS 2024 · 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.
