Research map / Reinforcement learning / RL from feedback
Adaptive Prior-Dependent Correction Enhanced Reinforcement Learning for Natural Language 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 |
|---|---|---|---|
| Add APDC to APG (ablation) | Transformer NMT · WMT14 En-De | BLEU: 27.63 -> 28.03 | Helped |
| Add APDC to APG (ablation) | Transformer NMT · WMT17 En-Zh | BLEU: 34.54 -> 34.91 | Helped |
| Add APDC to APG (ablation) | Transformer NMT · WMT17 Zh-En | BLEU: 24.81 -> 25.28 | Helped |
| Add APDC to APG (ablation) | Top-down image captioner · MSCOCO Karpathy test split | CIDEr: 120.2 -> 123.6 | Helped |
| Add APDC to APG (ablation) | Pointer-Generator+Coverage summarizer · CNN/Daily Mail | ROUGE-1: 41.51 -> 42.73 | Helped |
| Add PDC without the adaptive factor (ablation) | APG for NMT · WMT14 En-De | BLEU: 27.63 -> 27.70 | Helped |
| Add PDC without the adaptive factor (ablation) | APG for image captioning · MSCOCO Karpathy test split | BLEU-4: 36.4 -> 36.7 | Helped |
| Add PDC without the adaptive factor (ablation) | APG for summarization · CNN/Daily Mail | ROUGE-1: 41.51 -> 41.60 | Helped |
| Add the adaptive factor to PDC (ablation) | APG for image captioning · MSCOCO Karpathy test split | BLEU-4: 36.7 -> 37.9 | Helped |
| Use METEOR rather than CIDEr as the reward (ablation) | APG+APDC image captioner · MSCOCO Karpathy test split | METEOR: 28.9 -> 29.2 | Helped |
| Use METEOR rather than CIDEr as the reward (ablation) | APG+APDC image captioner · MSCOCO Karpathy test split | CIDEr: 123.6 -> 119.7 | Hurt |
| Use the advantage function estimated from successive Q-values instead of other policy-gradient weighting | NLG policy-gradient training · NLG tasks | task evaluation metrics | Helped |
Closest papers by meaning
- Training Language Models to Self-Correct via Reinforcement Learning (ICLR 2025 · similarity 0.87)
- Reducing the Probability of Undesirable Outputs in Language Models Using Probabilistic Inference (NeurIPS 2025 · similarity 0.87)
- Towards Cost-Effective Reward Guided Text Generation (ICML 2025 · similarity 0.87)
- Degeneration-free Policy Optimization: RL Fine-Tuning for Language Models without Degeneration (ICML 2024 · similarity 0.86)
- Leftover Lunch: Advantage-based Offline Reinforcement Learning for Language Models (ICLR 2024 · similarity 0.86)
- Reinforcement Learning for Diffusion LLMs with Entropy-Guided Step Selection and Stepwise Advantages (NeurIPS 2026 · similarity 0.86)
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.