Research map / Reinforcement learning / RL from feedback
RESTL: Reinforcement Learning Guided by Multi-Aspect Rewards for Signal Temporal Logic Transformation
Yue Fang, Zhi Jin, Jie An, Hongshen Chen, Xiaohong Chen, Naijun Zhan
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 |
|---|---|---|---|
| add atomic proposition alignment reward m_a (ablation) | RESTL · DeepSTL | formula accuracy: 0.5571 -> 0.5985 | Improved |
| add templated NL similarity reward m_t (ablation) | RESTL · STL-DivEn | formula accuracy: 0.6424 -> 0.6838 | Improved |
| add STL-level similarity reward m_s (ablation) | RESTL · STL-DivEn | formula accuracy: 0.6314 -> 0.6838 | Improved |
| add formula succinctness reward m_l (ablation) | RESTL · STL-DivEn | formula accuracy: 0.6642 -> 0.6838 | Improved |
| add formula succinctness reward m_l (ablation) | RESTL · DeepSTL | formula accuracy: 0.5832 -> 0.5985 | Improved |
| apply RL with multi-aspect reward models on top of fine-tuned LLaMA 3-8B (ablation) | LLaMA 3-8B · STL-DivEn | formula accuracy: 0.4956 -> 0.6838 | Improved |
| apply RL with multi-aspect reward models on top of fine-tuned LLaMA 3-8B (ablation) | LLaMA 3-8B · DeepSTL | formula accuracy: 0.2850 -> 0.5985 | Improved |
| use a single reward only instead of all four (ablation) | RESTL · STL-DivEn | formula accuracy: 0.6838 -> 0.5952 (m_l only) | Hurt |
| train reward models with easy-to-hard curriculum instead of reverse order (ablation) | reward model r_a · STL-DivEn | reward model accuracy (%): 74.6 -> 80.1 | Improved |
| use reward models instead of direct metric supervision in RL (ablation) | RESTL · STL-DivEn | template accuracy (%): 60.3 -> 61.2 | Improved |
| use reward models instead of direct metric supervision in RL (ablation) | RESTL · STL-DivEn | BLEU: 0.267 -> 0.276 | Improved |
| RESTL vs KGST in human evaluation | RESTL · STL-DivEn human evaluation | win rate (%): win 58.7, loss 15.9 | Improved |
Closest papers by meaning
- Supervised Reinforcement Learning: From Expert Trajectories to Step-wise Reasoning (ICLR 2026 · similarity 0.91)
- DeepLTL: Learning to Efficiently Satisfy Complex LTL Specifications for Multi-Task RL (ICLR 2025 · similarity 0.88)
- Temporal Logic Specification-Conditioned Decision Transformer for Offline Safe Reinforcement Learning (ICML 2024 · similarity 0.88)
- RL Squeezes, SFT Expands: A Comparative Study of Reasoning LLMs (ICLR 2026 · similarity 0.88)
- From Verifiable Dot to Reward Chain: Harnessing Verifiable Reference-based Rewards for Reinforcement Learning of Open-ended Generation (ICLR 2026 · similarity 0.88)
- Resource-Efficient Reinforcement for Reasoning Large Language Models via Dynamic One-Shot Policy Refinement (ICML 2026 · 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.
