Research map / Large language models / Agents, code and math
Deep Reinforcement Learning for Syntactic Error Repair in Student Programs
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 expert demonstrations for one tenth of training programs plus small edit penalty (ablation) | RLAssist · DeepFix C student programs (6975 erroneous) | completely fixed programs: 1699 (24.4%) -> 1854 (26.6%) | Helped |
| add expert demonstrations and edit penalty (ablation) | RLAssist · DeepFix C student programs (6975 erroneous) | error messages resolved: 5884 (35.1%) -> 6652 (39.7%) | Helped |
| add expert demonstrations and edit penalty (ablation) | RLAssist · training | episodes until visible progress (millions): 1.2 -> 0.2 | Helped |
| add expert demonstrations and edit penalty (ablation) | RLAssist · training | training time | Helped |
| add edit penalty without demonstrations (ablation) | RLAssist · training | trainability | Hurt |
| use more demonstrations than one tenth (ablation) | RLAssist+Demo · training | training speed | No clear effect |
| use fewer demonstrations than one tenth (ablation) | RLAssist+Demo · training | training speed | Hurt |
| RL token-level agent without supervision instead of supervised NMT DeepFix | RLAssist · DeepFix C student programs (6975 erroneous) | completely fixed programs: 1625 (23.3%) -> 1699 (24.4%) | Helped |
Closest papers by meaning
- Training Language Models to Self-Correct via Reinforcement Learning (ICLR 2025 · similarity 0.87)
- $\mathcal{B}$-Coder: Value-Based Deep Reinforcement Learning for Program Synthesis (ICLR 2024 · similarity 0.87)
- Self-play with Execution Feedback: Improving Instruction-following Capabilities of Large Language Models (ICLR 2025 · similarity 0.86)
- Afterburner: Reinforcement Learning Facilitates Self-Improving Code Efficiency Optimization (NeurIPS 2025 · similarity 0.86)
- RLEF: Grounding Code LLMs in Execution Feedback with Reinforcement Learning (ICML 2025 · similarity 0.85)
- Counterexample Guided Program Repair Using Zero-Shot Learning and MaxSAT-based Fault Localization (AAAI 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.