Research map / Large language models / Agents, code and math
Learner-Tailored Program Repair: A Solution Generator with Iterative Edit-Driven Retrieval Enhancement
Zhenlong Dai, Zhuoluo Zhao, Hengning Wang, Xiu Tang, Sai Wu, Chang Yao, Zhipeng Gao, Jingyuan Chen
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 Iterative Retrieval Enhancement (3 iterations) to LSGen (ablation) | LSGen (GPT-4o) · LPR-Bench | Acc (%): 80.59 -> 91.40 | Improved |
| add edit-driven solution retrieval versus cosine code retrieval (ablation) | LSGen (GPT-4o) · LPR-Bench ablation | Acc (%): 86.00 -> 87.96 | Improved |
| add reference-inspired solution generation (bug description plus diff) (ablation) | LSGen (Qwen2.5-Coder-7B) · LPR-Bench ablation | Acc (%): 25.55 -> 51.11 | Improved |
| add reference-inspired solution generation (bug description plus diff) (ablation) | LSGen (GPT-4o) · LPR-Bench ablation | Acc (%): 77.40 -> 87.96 | Improved |
| add reference-inspired solution generation (ablation) | LSGen (Claude-4) · LPR-Bench ablation | B-F1 (%): 35.58 -> 37.10 | Mixed |
| add edit-driven retrieval on top of the rest of LSGen (ablation) | LSGen (Qwen2.5-Coder-7B) · LPR-Bench ablation | B-F1 (%): 16.96 -> 19.91 | Improved |
| replace baseline retrievers with LSGen on GPT-4o | LSGen (GPT-4o) · LPR-Bench | Acc (%): 40.54 -> 91.40 | Improved |
| increase iteration count from 1 to 3 (ablation) | LSGen (GPT-4o, Qwen3-Embedding-0.6B) · LPR-Bench | Acc (%): 87.96 -> 91.40 | Improved |
| increase iteration count from 1 to 3 (ablation) | LSGen (Qwen2.5-Coder-7B, UniXcoder) · LPR-Bench | Acc (%): 52.83 -> 58.23 | Improved |
| direct prompting with AdaPatcher two-stage localization over NoRef | GPT-4o · LPR-Bench | Acc (%): 18.43 -> 19.41 | No clear effect |
Closest papers by meaning
- Counterexample Guided Program Repair Using Zero-Shot Learning and MaxSAT-based Fault Localization (AAAI 2025 · similarity 0.90)
- Code Repair with LLMs gives an Exploration-Exploitation Tradeoff (NeurIPS 2024 · similarity 0.88)
- Less Is More: Adaptive Program Repair with Bug Localization and Preference Learning (AAAI 2025 · similarity 0.87)
- CoRNStack: High-Quality Contrastive Data for Better Code Retrieval and Reranking (ICLR 2025 · similarity 0.85)
- Is Self-Repair a Silver Bullet for Code Generation? (ICLR 2024 · similarity 0.85)
- AuPair: Golden Example Pairs for Code Repair (ICML 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.
