Research map / Probabilistic and scientific modeling / Variational and generative sampling
Reward Augmented Maximum Likelihood for Neural Structured Prediction
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 |
|---|---|---|---|
| RAML with tau=0.9 instead of ML training (ablation) | attention seq2seq · TIMIT dev | phone error rate: 20.87 -> 18.00 | Helped |
| RAML with tau=0.9 instead of ML training (ablation) | attention seq2seq · TIMIT test | phone error rate: 22.18 -> 19.89 | Helped |
| raise temperature from 0.6 to 0.9 (ablation) | RAML · TIMIT dev | phone error rate: 19.92 -> 18.00 | Helped |
| raise temperature beyond 0.9 to 1.0 (ablation) | RAML · TIMIT dev | phone error rate: 18.00 -> 18.78 | Hurt |
| raise temperature beyond 0.9 to 1.0 (ablation) | RAML · TIMIT test | phone error rate: 19.89 -> 20.41 | Hurt |
| lowest temperature tau=0.60 vs ML (ablation) | RAML · TIMIT test | phone error rate: 22.18 -> 21.65 | Helped |
| RAML tau=0.85 instead of ML training (ablation) | attention seq2seq · WMT'14 En-Fr | average BLEU: 36.50 -> 36.91 | Helped |
| RAML tau=0.85 instead of ML training (ablation) | attention seq2seq · WMT'14 En-Fr | best BLEU: 36.87 -> 37.23 | Helped |
| raise temperature from 0.75 to 0.85 (ablation) | RAML · WMT'14 En-Fr | average BLEU: 36.62 -> 36.91 | Helped |
| raise temperature from 0.85 to 0.95 (ablation) | RAML · WMT'14 En-Fr | average BLEU: 36.91 -> 36.57 | Hurt |
| tau=0.95 versus ML baseline (ablation) | RAML · WMT'14 En-Fr | best BLEU: 36.87 -> 36.94 | Helped |
Closest papers by meaning
- Portable Reward Tuning: Towards Reusable Fine-Tuning across Different Pretrained Models (ICML 2025 · similarity 0.85)
- LASeR: Learning to Adaptively Select Reward Models with Multi-Arm Bandits (NeurIPS 2025 · similarity 0.85)
- Active Reward Modeling: Adaptive Preference Labeling for Large Language Model Alignment (ICML 2025 · similarity 0.85)
- On the Limits of Test-Time Compute: Sequential Reward Filtering for Better Inference (ICML 2026 · similarity 0.85)
- GRAM: A Generative Foundation Reward Model for Reward Generalization (ICML 2025 · similarity 0.84)
- BRAIn: Bayesian Reward-conditioned Amortized Inference for natural language generation from feedback (ICML 2024 · 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.